Customer group classification method and device based on machine learning, and electronic equipment

Abstract

The invention discloses a customer group classification method and device based on machine learning and electronic equipment, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the dimension reduction processing of an original data set through employing a kernel principal component analysis algorithm, obtaining a data point set after the dimension reduction, calculating the outlier factor of each data point through employing a local density factor algorithm, and obtaining an outlier classification result; removing abnormal data values by using outlier factors, calculating connection strength between nodes according to a relationship between data points, establishing a linear equation set, solving the linear equation set, obtaining an initial clustering center point, performing clustering analysis based on the initial clustering center point, classifying each data point in a data point set after dimension reduction into a cluster to which the closest centroid belongs, and performing clustering analysis based on the clustering center point. And obtaining a financial management customer group classification result until a clustering convergence condition is met. The technical problems that the real-time updating demand of large-scale customer data cannot be adapted during financial management customer classification in the related technology, the recommended product does not meet the demand, and the customer satisfaction is reduced are solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, financial technology or other related fields, and specifically to a customer classification method and device based on machine learning, and an electronic device. Background Art

[0002] In the financial industry, particularly banking, the volume and complexity of customer data has increased dramatically with the advent of the digital age. Accurate customer segmentation is crucial for providing personalized services, risk management, and product promotion. Traditional customer segmentation methods, such as rule-based classification and simple statistical clustering, struggle to cope with the high dimensionality and nonlinear nature of modern data. Their limitations are particularly pronounced when dealing with heterogeneous, multi-source data, such as that of bank wealth management customers, which includes extensive transaction records, investment behavior, and credit assessments.

[0003] Furthermore, existing wealth management customer classification methods, such as rule-based classification and simple cluster analysis, often stratify or categorize customers based on pre-set business logic or statistical algorithms. These methods performed adequately in earlier environments with smaller data volumes and relatively simple data structures. However, faced with the explosive growth of banking data and increasingly complex customer behavior patterns, traditional customer classification techniques have significant drawbacks. First, they are inefficient when processing large datasets and consume significant computing resources, making it difficult to process and analyze the massive amounts of customer data continuously accumulating within banks in real time. This processing capability shortfall is particularly pronounced in wealth management service scenarios with high concurrency and high real-time requirements. Second, as data dimensionality increases, traditional classification algorithms struggle to capture subtle changes and potential correlations in customer behavior patterns. They also lack compatibility and flexibility in processing multi-source, heterogeneous data, making them unable to meet rapidly evolving market conditions and customer needs. Third, customer data in wealth management businesses is dynamic, requiring retraining of traditional customer classification models when data is updated. This is not only time-consuming but also prone to lags, failing to reflect the latest customer information, impacting the timeliness and accuracy of decision-making.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The embodiments of the present invention provide a customer classification method and device based on machine learning, and an electronic device, to at least solve the technical problem in related technologies that when classifying financial management customers, they cannot adapt to the real-time update requirements of large-scale customer data, and the recommended products do not meet the requirements, resulting in a decrease in customer satisfaction.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a customer classification method based on machine learning is provided, including: obtaining an original data set of financial management customers, and using a kernel principal component analysis algorithm to perform dimensionality reduction processing on the original data set to obtain a reduced-dimensionality data point set, wherein the reduced-dimensionality data point set includes multiple data points and data values ​​corresponding to each data point; using a local density factor algorithm to calculate the outlier factor of each data point, and removing abnormal data values ​​in the reduced-dimensionality data point set according to the outlier factor; calculating the connection strength between nodes according to the relationship between data points, establishing a linear equation group, and using a preset iterative strategy to solve the linear equation group until the convergence condition is met to obtain initial cluster center points, wherein each initial cluster center point serves as the centroid of the clustering operation; performing cluster analysis based on the initial cluster center points, and classifying each data point in the reduced-dimensionality data point set into the cluster to which the centroid belongs closest, until the clustering convergence condition is reached, and obtaining the financial management customer classification result.

[0007] Optionally, the step of using a kernel principal component analysis algorithm to perform dimensionality reduction processing on the original data set to obtain a data point set after dimensionality reduction includes: using a selected kernel function to calculate the kernel matrix of the preprocessed original data set; centering the kernel matrix, and performing eigendecomposition processing on the centralized kernel matrix to obtain a set of eigenvalues ​​and an eigenvector corresponding to each eigenvalue; normalizing the eigenvectors; sorting all the normalized eigenvectors from large to small, selecting the eigenvectors corresponding to the first P eigenvalues, and obtaining the data point set after dimensionality reduction, wherein P is a positive integer greater than 1.

[0008] Optionally, the step of using a local density factor algorithm to calculate the outlier factor of each of the data points includes: using the local density factor algorithm to obtain a sample set that is k-th closest to the data point to obtain a k-nearest neighbor set, where k is a positive integer; obtaining the k-th reachable distance between the data point and another data point; based on the size of the k-nearest neighbor set and the k-th reachable distance, calculating the local reachable density of key parameters associated with the data point; and calculating the outlier factor of the data point based on the local reachable density of the key parameters of the data point, the size of the k-nearest neighbor set, and the average value of the local reachable density of another data point.

[0009] Optionally, the step of removing abnormal data values ​​in the data point set after dimensionality reduction according to the outlier factor includes: when the absolute value of the difference between the outlier factor and the preset outlier threshold is greater than the preset absolute value threshold, confirming that the data point is an outlier, taking the data value corresponding to the data point as the abnormal data value, and removing the abnormal data value in the data point set after dimensionality reduction; when the absolute value of the difference between the outlier factor and the preset outlier threshold is less than or equal to the preset absolute value threshold, confirming that the data point belongs to a normal cluster and retaining the data value corresponding to the data point.

[0010] Optionally, the step of calculating the connection strength between nodes based on the relationship between data points and establishing a set of linear equations includes: taking each data point in the data point set after dimensionality reduction as a node to construct a directed connection graph; calculating the connection strength between nodes based on the distance between each node in the directed connection graph and a control parameter, wherein the control parameter is a pre-set rate for adjusting the rate at which the connection strength between data points changes with distance; and constructing the set of linear equations based on the calculated connection strength between nodes, the damping factor, the transfer matrix generated by the Markov chain traversing the nodes, and the cluster reference PageRank value of each node transmitted through the link relationship.

[0011] Optionally, the step of solving the linear equation group using a preset iterative strategy until the convergence condition is met to obtain the initial cluster center point includes: iteratively updating each vector element in the linear equation group using a preset iterative strategy; monitoring the change of the linear equation group during the iterative update process until the convergence condition is met, and outputting the cluster reference PageRank value of each node, wherein the convergence condition includes: the element centroid change parameter is less than a preset parameter change threshold or the maximum number of iterative updates is reached; sorting the cluster reference PageRank values ​​of all the nodes from large to small, and selecting the top F data points as the initial cluster center points, where F is a positive integer.

[0012] Optionally, each data point in the data point set after dimensionality reduction is assigned to the cluster to which the nearest centroid belongs until the clustering convergence condition is reached, and the step of obtaining the classification result of the financial management customer group includes: obtaining a pre-set centroid change threshold and a maximum number of clustering iterations; a data point assignment step, calculating the distance from each data point in the data point set after dimensionality reduction to each centroid, and assigning each data point to the cluster to which the nearest centroid belongs according to the distance; a new centroid selection step, recalculating the centroid of each cluster, and taking the mean of all data points in the current cluster as the new centroid; iteratively executing the data point assignment step and the new centroid selection step until the clustering convergence condition is reached, wherein the clustering convergence condition includes: the number of iterations reaches the maximum number of clustering iterations, the new centroid change value is less than the centroid change threshold, or the cluster assignment no longer changes; based on the cluster to which the centroid of each data point belongs determined by the last iterative execution, the classification result of the financial management customer group is determined.

[0013] Optionally, after obtaining the original data set of financial management customers, it also includes: performing missing value processing on the original data set, wherein the missing value processing includes: mean filling, mode filling and deleting invalid missing values; converting text information in the original data set into numerical information; and standardizing the sample data in the original data set based on the standard deviation of the original data.

[0014] According to another aspect of an embodiment of the present invention, a customer group classification device based on machine learning is also provided, including: a data dimension reduction unit, used to obtain an original data set of financial management customers, and use a kernel principal component analysis algorithm to reduce the dimension of the original data set to obtain a reduced-dimensional data point set, wherein the reduced-dimensional data point set includes multiple data points and data values ​​corresponding to each data point; an outlier removal unit, used to use a local density factor algorithm to calculate the outlier factor of each of the data points, and remove abnormal data values ​​in the reduced-dimensional data point set according to the outlier factor; a cluster center point determination unit, used to calculate the connection strength between nodes based on the relationship between data points, establish a linear equation system, and use a preset iterative strategy to solve the linear equation system until the convergence condition is met to obtain initial cluster center points, wherein each initial cluster center point serves as the centroid of the clustering operation; a customer group classification unit, used to perform cluster analysis based on the initial cluster center points, classify each data point in the reduced-dimensional data point set into a cluster to which the centroid belongs closest, until the cluster convergence condition is met, and obtain the financial management customer group classification result.

[0015] Optionally, the data dimensionality reduction unit includes: a kernel matrix calculation module, which is used to use a selected kernel function to calculate the kernel matrix of the preprocessed original data set; a centralization processing module, which is used to centralize the kernel matrix and perform eigendecomposition processing on the centralized kernel matrix to obtain a set of eigenvalues ​​and an eigenvector corresponding to each eigenvalue; a normalization processing module, which is used to normalize the eigenvectors; and an eigenvector selection module, which is used to sort all the normalized eigenvectors from large to small, select the eigenvectors corresponding to the first P eigenvalues, and obtain the data point set after dimensionality reduction, where P is a positive integer greater than 1.

[0016] Optionally, the outlier removal unit includes: a sample set acquisition module, which is used to use a local density factor algorithm to acquire a sample set that is closest to the data point to obtain a k-th nearest neighbor set, where k is a positive integer; obtain the k-th reachable distance between the data point and another data point; a local reachable density calculation module, which is used to calculate the local reachable density of key parameters associated with the data point based on the size of the k-th nearest neighbor set and the k-th reachable distance; and an outlier factor calculation module, which is used to calculate the outlier factor of the data point based on the local reachable density of key parameters of the data point, the size of the k-th nearest neighbor set, and the average value of the local reachable density of another data point.

[0017] Optionally, the outlier removal unit includes: an outlier confirmation module, which is used to confirm that a data point is an outlier when the absolute value of the difference between the outlier factor and the preset outlier threshold is greater than the preset absolute value threshold, and use the data value corresponding to the data point as the abnormal data value, and remove the abnormal data value in the data point set after dimensionality reduction; a normal data point confirmation module, which is used to confirm that a data point belongs to a normal cluster when the absolute value of the difference between the outlier factor and the preset outlier threshold is less than or equal to the preset absolute value threshold, and retain the data value corresponding to the data point.

[0018] Optionally, the cluster center determination unit includes: a directed graph construction module, which is used to use each data point in the data point set after dimensionality reduction as a node to construct a directed connection graph; a node connection strength calculation module, which is used to calculate the node connection strength based on the distance between each node in the directed connection graph and the control parameter, wherein the control parameter is a pre-set rate for adjusting the rate at which the connection strength between data points changes with distance; a linear equation group construction module, which is used to construct the linear equation group based on the calculated node connection strength, damping factor, transfer matrix generated by Markov chain traversal of nodes, and cluster reference PageRank value of each node transmitted through the link relationship.

[0019] Optionally, the cluster center determination unit also includes: an equation group iterative update module, which is used to iteratively update each vector element in the linear equation group using a preset iterative strategy; an update monitoring module, which is used to monitor the change of the linear equation group during the iterative update process until the convergence condition is met, and output the cluster reference PageRank value of each node, wherein the convergence condition includes: the element centroid change parameter is less than a preset parameter change threshold or the maximum number of iterative updates is reached; a cluster center selection module, which is used to sort the cluster reference PageRank values ​​of all the nodes from large to small, and select the top F data points as the initial cluster center points, where F is a positive integer.

[0020] Optionally, the customer group classification unit includes: an iterative parameter acquisition module for obtaining a pre-set centroid change threshold and a maximum number of clustering iterations; a data point attribution module for calculating the distance from each data point to each centroid in the data point set after dimensionality reduction, and assigning each data point to the cluster to which the nearest centroid belongs according to the distance; a new centroid selection module for recalculating the centroid of each cluster, and taking the mean of all data points in the current cluster as the new centroid; iteratively executing the data point attribution step and the new centroid selection step until the clustering convergence condition is reached, wherein the clustering convergence condition includes: the number of iterations reaches the maximum number of clustering iterations, the new centroid change value is less than the centroid change threshold, or the cluster assignment no longer changes; a customer group classification module for determining the financial management customer group classification result based on the cluster to which the centroid of each data point belongs determined by the last iterative execution.

[0021] Optionally, the customer classification device based on machine learning also includes: a missing value processing unit, which is used to perform missing value processing on the original data set after obtaining the original data set of financial management customers, wherein the missing value processing includes: mean filling, mode filling and deletion of invalid missing values; a numerical conversion unit, which is used to convert the text information in the original data set into numerical information; and a standardization unit, which is used to standardize the sample data in the original data set based on the standard deviation of the original data.

[0022] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned customer classification methods based on machine learning.

[0023] According to another aspect of an embodiment of the present invention, an electronic device is also provided, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the above-mentioned customer classification methods based on machine learning.

[0024] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned customer group classification methods based on machine learning.

[0025] In the present disclosure, an original data set of financial management customers can be obtained, and the kernel principal component analysis algorithm is used to perform dimensionality reduction processing on the original data set to obtain a data point set after dimensionality reduction, wherein the data point set after dimensionality reduction includes multiple data points and the data value corresponding to each data point; a local density factor algorithm is used to calculate the outlier factor of each data point, and abnormal data values ​​in the data point set after dimensionality reduction are removed according to the outlier factor; the connection strength between nodes is calculated according to the relationship between data points, a linear equation group is established, and a preset iterative strategy is used to solve the linear equation group until the convergence condition is met, and the initial cluster center point is obtained, wherein each initial cluster center point serves as the centroid of the clustering operation; cluster analysis is performed based on the initial cluster center point, and each data point in the data point set after dimensionality reduction is classified into the cluster to which the centroid belongs closest, until the cluster convergence condition is reached, and the classification result of the financial management customer group is obtained.

[0026] Based on the above-mentioned public content, the kernel principal component analysis algorithm can be used to reduce the dimensionality of the original data set, and the local density factor algorithm can be used to remove abnormal data in the data point set after dimensionality reduction. Then, the initial clustering center can be optimized to improve the clustering effect of the clustering algorithm. Through the final clustering results, the customer's financial management needs can be accurately identified, and more suitable financial management products can be recommended to financial customers, thereby solving the technical problem in related technologies that when classifying financial management customers, it is impossible to adapt to the real-time update requirements of large-scale customer data, and the recommended products do not meet the needs, resulting in a decline in customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0028] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a customer group classification method based on machine learning is shown;

[0029] Figure 2is a flowchart of an optional customer group classification method based on machine learning according to an embodiment of the present invention;

[0030] Figure 3 This is a flowchart of an optional method for classifying financial management customer groups based on kernel principal component analysis and an improved clustering algorithm according to an embodiment of the present invention;

[0031] Figure 4 is a schematic diagram of an optional customer group classification device based on machine learning according to an embodiment of the present invention;

[0032] Figure 5 This is a structural block diagram of an electronic device that executes a customer group classification method based on machine learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] To facilitate those skilled in the art to understand the present invention, some of the terms or nouns involved in the embodiments of the present invention are explained below:

[0036] Kernel Principal Component Analysis (KPCA) transforms data into a high-dimensional space, uses a kernel function to calculate the similarity between data points, and then performs principal component analysis in that high-dimensional space. KPCA preserves the nonlinear structure of the data while reducing its dimensionality, providing a more optimized dataset for subsequent cluster analysis. It is particularly effective in processing the complex nonlinear relationships in the wealth management customer data of financial institutions.

[0037] The Local Outlier Factor (LOF) algorithm is a density-based outlier detection algorithm used to identify points in a dataset that have a lower density than surrounding data points, known as outliers. It calculates the local density of each point and compares it with the average density of its neighboring points to determine whether it is an outlier. In this paper, LOF is used in the preprocessing stage to filter out outliers, preventing them from affecting the results of the clustering algorithm, thereby improving the accuracy and stability of customer classification.

[0038] PageRank Algorithm, abbreviated as PageRank, in the optimized clustering algorithm of the present invention, the PageRank algorithm is used to determine the relative importance of data points in the network, and then select the point that best represents the data distribution as the initial clustering center of the clustering algorithm, thereby improving the convergence speed and clustering effect of the clustering algorithm and avoiding the local optimal solution problem that may be caused by randomly selecting the initial centroid.

[0039] Gauss-Seidel Iteration Method, abbreviated as Gauss-Seidel, is an iterative algorithm for solving linear equations. In the present invention, the Gauss-Seidel Iteration Method is used to optimize the calculation process of the PageRank algorithm, improve the calculation speed of the PageRank value, reduce the calculation cost, and ensure the accuracy of the results.

[0040] The clustering algorithm in this invention optimizes the initial centroid through the PageRank algorithm, combines the LOF algorithm to filter outliers, and introduces a dynamic threshold mechanism and a maximum number of iterations strategy to achieve efficient and accurate classification of financial institution financial management customer groups, can adapt to the real-time update needs of large-scale data, and enhance the practicality and flexibility of customer classification.

[0041] It should be noted that the customer classification method and device based on machine learning in the present disclosure can be used in the field of artificial intelligence technology. In the case of financial institution financial management customer classification based on the kernel principal component analysis algorithm and the clustering strategy optimized by the PageRank algorithm, it can also be used in any field other than the field of artificial intelligence technology. In the case of financial institution financial management customer classification based on the kernel principal component analysis algorithm and the clustering strategy optimized by the PageRank algorithm, the application field of the customer classification method and device based on machine learning in the present disclosure is not limited.

[0042] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) collected by this disclosure are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation portals for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or institution through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or institution.

[0043] It should be noted that in this disclosure, when collecting and analyzing customer information, the corresponding operation entrance is provided for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered.

[0044] The following embodiments of the present invention can be applied to various systems / applications / devices for customer classification based on machine learning. The present invention can be applied to the refined management and service provision scenarios of financial institutions' wealth management customer groups. In the financial management business of financial institutions, the present invention can conduct in-depth analysis of customer data, establish a more accurate customer classification model, and help financial institutions identify customer groups with different wealth management needs and risk tolerance, thereby providing more personalized wealth management products and services. In addition, the present invention can reveal potential customer risk patterns through efficient dimensionality reduction and clustering of customer data, provide data support for banks' risk management and credit decisions, reduce credit risks, and optimize resource allocation.

[0045] Based on accurate customer classification, financial institutions can provide more personalized financial advice, product recommendations and risk assessments, enhancing customer experience and improving the market competitiveness of financial institutions' services.

[0046] The present invention will be described in detail below with reference to various embodiments.

[0047] Example 1

[0048] According to an embodiment of the present invention, an embodiment of a customer classification method based on machine learning is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0049] The customer group classification method embodiment based on machine learning provided in Example 1 of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a customer group classification method based on machine learning is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more ( Figure 1 The computer system includes a processor 102 (shown as 102a, 102b, ..., 102n) (the processor 102 may include but is not limited to a microcontroller unit (MCU) or a programmable logic device (FPGA)), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, the computer system may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS), a network interface, a power supply, and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0050] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0051] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the customer group classification method based on machine learning in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned customer group classification method based on machine learning. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0052] The transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0053] The display may be, for example, a touch screen liquid crystal display (LCD), which enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0054] Under the above operating environment, this application provides Figure 2 The customer classification method based on machine learning is shown. Figure 2 is a flow chart of an optional customer group classification method based on machine learning according to an embodiment of the present invention, such as Figure 2 As shown, the method includes the following steps:

[0055] Step S201: obtain the original data set of the financial management customer, and use the kernel principal component analysis algorithm to reduce the dimension of the original data set to obtain a data point set after dimensionality reduction, wherein the data point set after dimensionality reduction includes multiple data points and the data value corresponding to each data point.

[0056] This example first collects a dataset containing all relevant customer information, including but not limited to basic customer information, financial transaction history, financial product preferences, risk tolerance, and other multi-dimensional data. This data is sourced from a wide range of sources, including internal bank systems, third-party data providers, social media, and other channels, ensuring comprehensiveness and diversity.

[0057] The acquired data set may contain data redundancy, data missing, etc. Therefore, before officially entering the dimensionality reduction process, this embodiment will perform a series of preprocessing on the data. After obtaining the original data set of financial management customers, this embodiment needs to preprocess the data. Optionally, after obtaining the original data set of financial management customers, the process also includes: performing missing value processing on the original data set, wherein missing value processing includes: mean filling, mode filling, and deleting invalid missing values; converting text information in the original data set into numerical information; and standardizing the sample data in the original data set based on the standard deviation of the original data.

[0058] For continuous numerical features, if missing values ​​exist, this embodiment uses a mean filling strategy, replacing the missing value with the average of all non-missing data for that feature. For discrete or categorical features, a mode filling method is used, where the mode refers to the most frequently occurring value in the dataset. When performing mode filling, this embodiment selects the mode of the feature to fill the missing value, which can reflect the most common category state in the dataset and maintain the representativeness of the categorical feature.

[0059] In addition, in some cases, there are so many missing values ​​in some columns or rows in the data set that the filling strategy cannot provide a reasonable estimate. In this case, this embodiment may choose to directly delete these invalid columns or rows to avoid data skew or incorrect analysis caused by a large number of missing values.

[0060] It should be noted that this embodiment takes into account that the original data set may contain text information that is crucial to customer classification, such as job descriptions, investment preference statements, etc. In order to integrate this text information into the data analysis process, information extraction and conversion technology is adopted. First, the text information can be pre-processed using natural language processing (NLP) technology, such as word segmentation, stop word removal, stemming, and other steps to convert the text into a structured word list or vector representation. Subsequently, word frequency-inverse document frequency, word embedding, or other advanced text feature extraction methods can be used to convert the text information into numerical information so that it can participate in subsequent analysis and modeling together with other numerical features.

[0061] During the data preprocessing stage, this embodiment also performs standardization on the sample data in the original data set. For example, the Z-Score standardization method is used to convert the data value of each feature into its degree of deviation from the feature mean, and measured by the standard deviation. The values ​​of all features will fall within the same scale range, eliminating the impact of differences in feature dimensions or magnitudes, making the data dimensionality reduction and clustering analysis process more fair and reasonable, and not affected by the feature scale.

[0062] Optionally, the steps of using a kernel principal component analysis algorithm to perform dimensionality reduction processing on the original data set to obtain a data point set after dimensionality reduction include: using a selected kernel function to calculate the kernel matrix of the preprocessed original data set; centering the kernel matrix, and performing eigendecomposition processing on the centralized kernel matrix to obtain a set of eigenvalues ​​and an eigenvector corresponding to each eigenvalue; normalizing the eigenvectors; sorting all the normalized eigenvectors from large to small, selecting the eigenvectors corresponding to the first P eigenvalues, and obtaining a data point set after dimensionality reduction, where P is a positive integer greater than 1.

[0063] After completing data preprocessing, this embodiment uses the kernel principal component analysis algorithm to perform dimensionality reduction on the preprocessed original data set. First, an appropriate kernel function (such as a Gaussian kernel, a linear kernel, or a polynomial kernel) is selected, and the kernel matrix of the data set is calculated based on this kernel function to capture the nonlinear relationship between the data. Then, the obtained kernel matrix is ​​centralized to eliminate data bias and enhance the stability of the algorithm. After that, the centralized kernel matrix is ​​eigendecomposed to obtain a set of eigenvalues ​​and corresponding eigenvectors. The eigenvectors are normalized to ensure that the length of each eigenvector is unit length to avoid the scale effect affecting the analysis results.

[0064] After feature decomposition, this embodiment further sorts all normalized eigenvectors according to the size of their corresponding eigenvalues. The eigenvalue reflects the weight of the eigenvector in the dimensionality reduction space. Therefore, this embodiment selects the eigenvectors corresponding to the first P largest eigenvalues ​​(P is a positive integer greater than 1) to construct the data representation space after dimensionality reduction. This can retain the intrinsic structure and information of the data set to the greatest extent, while reducing the data dimension, reducing the computational complexity of subsequent clustering analysis, and improving efficiency.

[0065] Step S202 : using a local density factor algorithm to calculate the outlier factor of each data point, and removing abnormal data values ​​in the data point set after dimensionality reduction according to the outlier factor.

[0066] Optionally, the step of using a local density factor algorithm to calculate the outlier factor of each data point includes: using the local density factor algorithm to obtain a sample set that is k-th closest to the data point to obtain a k-nearest neighbor set, where k is a positive integer; obtaining the k-th reachable distance between the data point and another data point; based on the size of the k-nearest neighbor set and the k-th reachable distance, calculating the local reachable density of key parameters of the associated data point; and calculating the outlier factor of the data point based on the local reachable density of the key parameters of the data point, the size of the k-nearest neighbor set, and the average value of the local reachable density of another data point.

[0067] In the technical process of this embodiment, step S202 uses the local density factor (LOF) algorithm to calculate the outlier factor of each data point after dimensionality reduction, and identifies and excludes abnormal data values ​​that may have a negative impact on subsequent cluster analysis. First, this embodiment determines the set of k nearest neighbor data points for each data point, where k is a positive integer representing the size of the neighborhood. Its selection needs to be weighed based on the characteristics of the data set and the expected effect of clustering. By calculating the distance between data points (usually using Euclidean distance), the k points closest to the current point are found to form a k-nearest neighbor set. After that, the k-th reachable distance between the data point and another data point can be obtained. The k-th reachable distance refers to the distance from the data point to its k-th nearest neighbor point among all the k-nearest neighbor points of the data point. Finally, based on the size of the k-nearest neighbor set and the k-th reachable distance, the key parameter local reachable density of the associated data point is calculated. Specifically, for each data point, this embodiment calculates the average k-reachable distance of its k-nearest neighbor points, and then calculates the local reachable density based on this distance.

[0068] Furthermore, when calculating the outlier factor (LOF), this embodiment needs to know the size of the data point and its k-nearest neighbor set, as well as the local reachable density of these neighborhood points themselves, that is, it is necessary to obtain the three-dimensional size of the k-nearest neighbor set of the data point and the average value of the local reachable density of another data point. Finally, based on the key parameters of the data point, the local reachable density, the size of the k-nearest neighbor set, and the average value of the local reachable density of another data point, the outlier factor of the data point is calculated. The outlier factor is an indicator for evaluating whether a data point is an outlier. The larger the LOF value, the higher the degree of outlier the data point.

[0069] By calculating the LOF value for each data point, this embodiment can identify data points with significantly lower density than the surrounding area, namely potential outliers or outliers. Then, based on the comparison between the outlier factor and the preset outlier threshold, it is determined whether a data point should be considered an outlier. Removing outliers can effectively optimize clustering results, preventing outliers from distorting the cluster center, and thus achieving more accurate customer segmentation.

[0070] Optionally, the step of removing abnormal data values ​​in the data point set after dimensionality reduction according to the outlier factor includes: when the absolute value of the difference between the outlier factor and the preset outlier threshold is greater than the preset absolute value threshold, confirming that the data point is an outlier, taking the data value corresponding to the data point as the abnormal data value, and removing the abnormal data values ​​in the data point set after dimensionality reduction; when the absolute value of the difference between the outlier factor and the preset outlier threshold is less than or equal to the preset absolute value threshold, confirming that the data point belongs to a normal cluster and retaining the data value corresponding to the data point.

[0071] This embodiment first calculates the LOF value of each data point. The size of the LOF value reflects the possibility of the data point being an outlier. Subsequently, these LOF values ​​are compared with a pre-set outlier threshold. If the absolute value of the difference between the LOF value of a data point and the preset outlier threshold is greater than the preset absolute value threshold, the data point is marked as an outlier. Once a data point is confirmed to be an outlier, this embodiment will mark the data value corresponding to the data point as an abnormal data value and remove it from the data point set after dimensionality reduction. If the absolute value of the difference between the LOF value of a data point and the preset outlier threshold is less than or equal to AVT, this means that the LOF value of the data point is consistent with or close to the normal performance of the cluster. Therefore, this embodiment regards it as part of a normal cluster and retains the data value corresponding to the data point.

[0072] In step S203, the connection strength between nodes is calculated based on the relationship between data points, a linear equation group is established, and the linear equation group is solved using a preset iterative strategy until the convergence condition is met to obtain the initial cluster center points, where each initial cluster center point serves as the centroid of the clustering operation.

[0073] Optionally, the step of calculating the connection strength between nodes based on the relationship between data points and establishing a set of linear equations includes: taking each data point in the data point set after dimensionality reduction as a node to construct a directed connection graph; calculating the connection strength between nodes based on the distance between each node in the directed connection graph and a control parameter, wherein the control parameter is a pre-set rate for adjusting the rate at which the connection strength between data points changes with distance; and constructing a set of linear equations based on the calculated connection strength between nodes, the damping factor, the transfer matrix generated by the Markov chain traversing the nodes, and the cluster reference PageRank value of each node transmitted through the link relationship.

[0074] In this step, this embodiment regards the data points after dimensionality reduction as nodes in the graph, and constructs a directed connection graph by considering the relationship between the data points, in which each directed edge represents a possible connection or influence between two data points. The construction of this graph structure provides a basis for the subsequent calculation of the connection strength between nodes, and also enables the use of the principles of the PageRank algorithm to optimize the initial centroid selection of the clustering algorithm.

[0075] This embodiment further quantifies the connection strength between nodes based on the distance between nodes in the directed connection graph, combined with preset control parameters. The function of the control parameters is to adjust the rate at which the connection strength between data points changes with distance, ensuring that the calculation of the connection strength not only reflects the proximity of the data points but also takes into account the overall distribution characteristics of the data set, so as to more reasonably construct the linear equation system.

[0076] Based on the calculated inter-node connection strengths and damping factors, as well as the transition matrix generated by traversing the nodes through a Markov chain, this embodiment constructs a linear equation system for PageRank. The PageRank value of each node represents the relative importance of the node in the network. This value is obtained by iteratively solving the linear equation system, ensuring that the selection of the centroid is based not only on the distribution of the data but also on the structural position of the data point in the network.

[0077] Optionally, the step of using a preset iterative strategy to solve the linear equations until the convergence condition is met to obtain the initial cluster center point includes: iteratively updating each vector element in the linear equations using the preset iterative strategy; monitoring the change of the linear equations during the iterative update process until the convergence condition is met, and outputting the cluster reference PageRank value of each node, wherein the convergence condition includes: the element centroid change parameter is less than the preset parameter change threshold or the maximum number of iterative updates is reached; the cluster reference PageRank values ​​of all nodes are sorted from large to small, and the top F data points are selected as the initial cluster center points, where F is a positive integer.

[0078] The preset iteration strategy can be selected at will. For example, the Gauss-Seidel iteration method is used to solve the linear equations constructed by the PageRank algorithm. In this embodiment, a vector is first initialized to represent the estimated PageRank value of each node, which is usually initialized to the same value to ensure fairness and initial consistency of the iteration. In each iteration, the PageRank value of each node is calculated and updated based on the current state of the linear equations and the results of the previous iteration.

[0079] It should be noted that during the iteration process, this embodiment continuously monitors the element centroid change parameter, that is, the change of the node's PageRank value between iterations, to determine whether the iteration is close to the final solution and avoid unnecessary calculations.

[0080] This embodiment sets two types of convergence conditions for the solution process of the linear equations. The first is to set a change threshold. When the change in the PageRank value of all nodes is less than this threshold, the iterative process is determined to have converged, the iteration is stopped and the current PageRank value is output as the result. The second is that this embodiment sets a maximum number of iterations. If the iteration reaches this upper limit, the iterative process will stop regardless of whether the change is less than the threshold to prevent it from falling into an endless loop. Once the convergence conditions are met, this embodiment will output the PageRank value of each node and sort them from large to small according to these values. Based on the sorting results, the top K (K is a positive integer representing the preset number of clusters) nodes are selected as the initial cluster center points, ensuring that the selection of the initial cluster center points takes into account both the central position of the data point in the network and the diversity of data distribution.

[0081] In this embodiment, by deeply integrating the PageRank algorithm with the clustering algorithm, not only do we overcome the blindness of traditional clustering algorithms in selecting initial centroids and their tendency to fall into local optimal solutions, we also fully utilize the inherent connections between data points in the network structure, enhancing the accuracy and robustness of cluster analysis. This comprehensive solution, combining complex network analysis with statistical clustering techniques, can significantly improve the efficiency and effectiveness of customer classification, especially when processing large-scale, multi-dimensional data on the wealth management customer base of financial institutions.

[0082] Step S204: perform cluster analysis based on the initial cluster center point, and classify each data point in the dimensionality-reduced data point set into the cluster to which the nearest centroid belongs until the cluster convergence condition is reached, thereby obtaining the classification result of the financial management customer group.

[0083] In this embodiment, step S204 may perform cluster analysis using an optimized clustering algorithm to divide the data point set after dimensionality reduction into a number of clusters with high similarity to form a classification result of the financial management customer group.

[0084] Optionally, each data point in the data point set after dimensionality reduction is assigned to the cluster to which the nearest centroid belongs until the clustering convergence condition is reached, and the steps for obtaining the classification results of the financial management customer group include: obtaining a pre-set centroid change threshold and a maximum number of clustering iterations; a data point assignment step, calculating the distance from each data point in the data point set after dimensionality reduction to each centroid, and assigning each data point to the cluster to which the nearest centroid belongs according to the distance; a new centroid selection step, recalculating the centroid of each cluster, and taking the mean of all data points in the current cluster as the new centroid; iteratively executing the data point assignment step and the new centroid selection step until the clustering convergence condition is reached, wherein the clustering convergence condition includes: the number of iterations reaches the maximum number of clustering iterations, the new centroid change value is less than the centroid change threshold, or the cluster assignment no longer changes; based on the cluster to which the centroid of each data point belongs determined by the last iterative execution, the classification results of the financial management customer group are determined.

[0085] Before starting cluster analysis, this embodiment sets two important parameters: Centroid Change Threshold (CCT) and Maximum Clustering Iteration (MCI). CCT is used to determine whether the centroid has changed significantly, while MCI limits the maximum number of algorithm iterations to prevent infinite loops.

[0086] This embodiment first performs the data point attribution step, that is, calculates the distance from each data point to all centroids in the data point set after dimensionality reduction, and assigns the data point to the cluster represented by the nearest centroid based on the principle of closest distance. This step is based on the existing initial cluster center point, and determines the cluster attribution of the data point through a distance measurement mechanism (usually using Euclidean Distance, ED), providing preliminary data point allocation for subsequent refined clustering. After the data point attribution step, this embodiment will recalculate the centroid of each cluster. Specifically, the position of the centroid can be updated by calculating the average value of the feature vectors of all data points in the cluster to complete the new centroid selection step. The calculation of the new centroid helps to adjust the center position of the cluster to more accurately reflect the central trend of the data points in the cluster and enhance the effect and stability of clustering.

[0087] This embodiment iteratively executes the data point attribution step and the new centroid selection step until the clustering convergence condition is met. The clustering convergence conditions include but are not limited to the following situations: First, the number of iterations reaches the maximum clustering iteration number (MCI): When the number of iterations reaches the pre-set maximum value, the clustering process will stop, even if the clustering result may not have reached the optimal value at this time, but this provides a hard condition for the algorithm to stop and prevent endless consumption of computing resources. Second, the new centroid change value is less than the centroid change threshold (CCT): In the continuous iteration process, if the moving distance of all centroids is less than the preset centroid change threshold, it is determined that the clustering process has converged and the iteration is stopped. Third, the cluster allocation no longer changes: Another sign that the cluster analysis has reached a stable state is that the cluster allocation of data points no longer changes in continuous iterations, which indicates that the data points have found the most suitable cluster and no further adjustment is required.

[0088] When the clustering process meets any of the above convergence conditions, this embodiment determines the final classification result of the financial management customer group based on the data point attribution results determined in the last iteration, that is, the centroid cluster to which each data point belongs. Through a series of clustering analyses, the data point set is effectively divided into several clusters with similar characteristics. Each cluster represents a specific financial management customer group category, providing refined data support for subsequent customer insights, product recommendations and risk management.

[0089] The cluster analysis process of this embodiment can gradually approach the optimal clustering state by continuously adjusting the cluster affiliation and centroid position of data points until the convergence condition is reached, thereby ensuring the high quality of the final customer classification results. It can not only help financial institutions better understand customer characteristics, but also guide product innovation and customization of personalized services. Through precise customer segmentation, financial institutions can provide financial products and services that are more in line with customer needs, improve customer satisfaction and loyalty, and at the same time optimize risk management strategies and improve overall operational efficiency.

[0090] Through the above steps, the original data set of financial management customers can be obtained, and the kernel principal component analysis algorithm is used to reduce the dimension of the original data set to obtain a data point set after dimensionality reduction, wherein the data point set after dimensionality reduction includes multiple data points and the data value corresponding to each data point; the local density factor algorithm is used to calculate the outlier factor of each data point, and the abnormal data values ​​in the data point set after dimensionality reduction are removed according to the outlier factor; the connection strength between nodes is calculated according to the relationship between data points, a linear equation group is established, and the linear equation group is solved using a preset iterative strategy until the convergence condition is met, and the initial cluster center point is obtained, wherein each initial cluster center point is used as the centroid of the clustering operation; cluster analysis is performed based on the initial cluster center point, and each data point in the data point set after dimensionality reduction is classified into the cluster to which the nearest centroid belongs until the cluster convergence condition is met, and the classification result of the financial management customer group is obtained. In this embodiment, the kernel principal component analysis algorithm can be used to reduce the dimensionality of the original data set, and the local density factor algorithm can be used to remove abnormal data in the data point set after dimensionality reduction. Then, the initial cluster center is optimized to improve the clustering effect of the clustering algorithm. The final clustering results can accurately identify the customer's financial management needs and recommend more suitable financial management products to financial customers, thereby solving the technical problem in related technologies that when classifying financial management customers, it is unable to adapt to the real-time update requirements of large-scale customer data, and the recommended products do not meet the needs, resulting in a decline in customer satisfaction.

[0091] The following describes in detail another optional specific implementation.

[0092] Figure 3 This is a flowchart of an optional method for classifying financial management customer groups based on kernel principal component analysis and an improved clustering algorithm according to an embodiment of the present invention. Figure 3 Shown, including:

[0093] Step 1: Data preprocessing.

[0094] 1. Fill in missing values.

[0095] In response to the missing values ​​that may exist in the original data sets of financial customers obtained by financial institutions, the embodiments of the present invention can perform operations such as mode filling, mean filling, and deletion of missing values. The specific usage method depends on the data type of the column where the missing values ​​are located, and the processing of missing data in combination with business requirements.

[0096] 2. Text information preprocessing.

[0097] For text information, the embodiment of the present invention first performs text preprocessing on it, converting the original text into clean and structured data, and uses operations such as word segmentation and feature extraction to convert the text into numerical information.

[0098] 3. Data standardization.

[0099] Optionally, the embodiment of the present invention may use the Z-Score normalization method, for the sample sequence x1, x2, ..., x n , Z-Score standardization has:

[0100]

[0101] Among them, x i is the original data, y i is the new data after standardization, and s is the standard deviation of the original data.

[0102] Step 2: KPCA algorithm implementation.

[0103] 1. Select the kernel function k(x i ,x j ), calculate its kernel matrix K.

[0104] 2. Centralize the kernel matrix K:

[0105]

[0106] Among them, 1 n is an n×n matrix, and all elements in the matrix are

[0107] 3. Centralized kernel matrix Perform eigendecomposition:

[0108]

[0109] The solution is eigenvalue λ1≥λ2≥...≥λ n , and the corresponding eigenvectors α1,α2,...,α n .

[0110] 4. To ensure that the modulus of the principal component is 1, normalize the eigenvector:

[0111]

[0112] 5. Select the eigenvectors α1,...,α corresponding to the first p largest eigenvalues p , and get the data point set after dimensionality reduction:

[0113]

[0114] Among them, z i ∈R p is the data point after dimensionality reduction.

[0115] Step 3: Implementation of PageRank algorithm based on Gauss-Seidel iteration.

[0116] 1. The core step of detecting abnormal data based on the Local Outlier Factor (LOF) algorithm is to calculate the outlier factor of each data point:

[0117]

[0118] Among them, |N k (x i )| is x i The size of the k-nearest neighbor set, Ird k (x j ) is point x j The local reachable density of Then x j The average of the local reachability densities.

[0119] When the LOF value is significantly higher than 1 (corresponding to the preset absolute value threshold mentioned above), the probability that the corresponding data object belongs to an outlier increases accordingly; conversely, the closer it is to 1, the higher the probability that the data point belongs to a normal cluster. The key parameter is the local reachable density Ird k (x i ) is expressed as:

[0120]

[0121] Among them, N k (x i ) is the data object x i The kth distance neighborhood of x i The set of samples closest to k; d(x i ,x j ) is x i ,x j distance, reach-dist k (x i ,x j ) is x i ,x j The k-reachable distance between them can be written as:

[0122] reach-dist k (x i ,x j )=max{k-distiance(x j ),d(x i ,x j )}.

[0123] 2. Input consists of n m-dimensional sample sets X={x1,x2,...,x n}, threshold ε, initial vector x (0) , and generate a directed graph based on the sample set (corresponding to the directed connection graph mentioned above): n and m are determined by the characteristics of the financial management customer base, where n is the number of customer samples and m is the number of characteristics possessed by the sample.

[0124] 3. Calculate the connection strength d(x i ,x j ):

[0125]

[0126] in, is the Euclidean distance from each node i to node j, and γ is the control parameter.

[0127] 4. Calculate the ratio of the connection strength between nodes i and j to the total connection strength of node i, and establish a linear equation system about PR

[0128]

[0129] The linear equations are:

[0130]

[0131] Among them, PR is the PageRank value of the directed graph, d is the damping factor, which is usually 0.85, and L refers to l(x i ,x j ) is the transfer matrix generated by the Markov chain traversing the node (corresponding to the transfer matrix generated by the Markov chain traversing the node above), R t Each component PR(x i ) is represented by the PageRank value of each node transferred through the link relationship (corresponding to the cluster reference PageRank value of each node transferred through the link relationship mentioned above), and satisfies:

[0132]

[0133] 5. Use the Gauss-Seidel iteration method to solve the linear equations of PR and update PR(x i ):

[0134]

[0135] This method improves convergence efficiency by dynamically updating the solved components, and the matrix block calculation strategy effectively reduces memory consumption.

[0136] 6. Continue iterating until you meet the ‖PR (k+1) -PR (k) <ε‖ until the threshold or the maximum number of iterations is reached, and the PR (x i )value.

[0137] 7. PR(x i ) values ​​are arranged in descending order, and the Top-k samples are selected as the initial cluster centers of the clustering algorithm.

[0138] Step 4: Implementation of clustering algorithm.

[0139] 1. Determine F initial cluster centers according to the PageRank algorithm and set the parameters for determining the end of the iteration: centroid change threshold ∈, maximum number of cluster iterations T max By setting a reasonable centroid change threshold and maximum number of iterations, this application can ensure that the clustering process reaches the expected convergence state in a shorter time, avoiding long iterative cycles and greatly saving computing resources.

[0140] 2. For each sample x in the dataset i Calculate its distance to the F centroids and assign it to the nearest cluster:

[0141]

[0142] Among them, c i is the cluster label of the i-th data point, μ j is the centroid of the jth cluster. (1≤j≤k).

[0143] 3. Recalculate the centroid of each cluster and take the mean of all data points in the current cluster as the new centroid:

[0144]

[0145] Among them, C j is the set of points in cluster j, |C j | is the number of points in cluster j.

[0146] Repeat step 2.3 until the centroid change is less than the threshold, the cluster assignment does not change, or the maximum number of iterations (T max ).

[0147] The wealth management customer classification method provided by the above-mentioned embodiment can extract the main features of the data by using the KPCA algorithm when the data structure is unknown or relatively complex, filter out abnormal values ​​by using the LOF algorithm, and process and calculate the initial cluster center value of the clustering algorithm by using the Gauss-Seidel iterative PageRank algorithm to improve the clustering effect of the clustering algorithm. It can accurately identify the wealth management needs of customers and recommend more suitable financial products to customers. Combining KPCA and the improved clustering algorithm, this embodiment can achieve refined classification of bank wealth management customer groups, identify small-scale customer groups with high similarity, provide financial institutions with a more accurate customer stratification view, facilitate the design of personalized products and services, and improve customer satisfaction.

[0148] This implementation method uses machine learning algorithms to dynamically improve classification models, which not only ensures the accuracy of customer portraits, but also greatly improves the efficiency and accuracy of financial institutions in customer value mining and risk management.

[0149] The following describes it in detail with reference to another embodiment.

[0150] Example 2

[0151] A customer group classification device based on machine learning provided in this embodiment includes multiple implementation units, each implementation unit corresponds to each implementation step in the above-mentioned embodiment 1. Its specific implementation method and beneficial effects can be referred to the above-mentioned method embodiment and will not be repeated here.

[0152] Figure 4 is a schematic diagram of an optional customer group classification device based on machine learning according to an embodiment of the present invention, such as Figure 4 As shown, the customer group classification device based on machine learning may include: a data dimension reduction unit 41, an outlier removal unit 42, a cluster center point determination unit 43, and a customer group classification unit 44.

[0153] Among them, the data dimensionality reduction unit 41 is used to obtain the original data set of financial management customers, and use the kernel principal component analysis algorithm to reduce the dimensionality of the original data set to obtain a reduced-dimensional data point set, wherein the reduced-dimensional data point set includes multiple data points and the data value corresponding to each data point.

[0154] The outlier removal unit 42 is configured to calculate the outlier factor of each data point using a local density factor algorithm, and remove outlier data values ​​from the data point set after dimensionality reduction according to the outlier factor.

[0155] The cluster center determination unit 43 is used to calculate the connection strength between nodes based on the relationship between data points, establish a linear equation system, and use a preset iterative strategy to solve the linear equation system until the convergence condition is met, thereby obtaining the initial cluster center points, wherein each initial cluster center point serves as the centroid of the clustering operation.

[0156] The customer group classification unit 44 is used to perform cluster analysis based on the initial cluster center point, and classify each data point in the dimensionality-reduced data point set into the cluster to which the nearest centroid belongs until the cluster convergence condition is reached, thereby obtaining the financial management customer group classification result.

[0157] The above-mentioned customer group classification device based on machine learning can obtain the original data set of financial management customers through the data dimensionality reduction unit 41, and use the kernel principal component analysis algorithm to reduce the dimensionality of the original data set to obtain a data point set after dimensionality reduction, wherein the data point set after dimensionality reduction includes multiple data points and the data value corresponding to each data point, and the outlier factor of each data point is calculated by the local density factor algorithm through the outlier removal unit 42, and the abnormal data values ​​in the data point set after dimensionality reduction are removed according to the outlier factor, and the connection strength between nodes is calculated according to the relationship between data points through the cluster center point determination unit 43, and a linear equation group is established. The linear equation group is solved using a preset iterative strategy until the convergence condition is met to obtain the initial cluster center point, wherein each initial cluster center point is used as the centroid of the clustering operation, and the customer group classification unit 44 performs cluster analysis based on the initial cluster center point, and each data point in the data point set after dimensionality reduction is classified into the cluster to which the centroid belongs closest until the cluster convergence condition is met, and the financial management customer group classification result is obtained. In this embodiment, the kernel principal component analysis algorithm can be used to reduce the dimensionality of the original data set, and the local density factor algorithm can be used to remove abnormal data in the data point set after dimensionality reduction. Then, the initial cluster center is optimized to improve the clustering effect of the clustering algorithm. The final clustering results can accurately identify the customer's financial management needs and recommend more suitable financial management products to financial customers, thereby solving the technical problem in related technologies that when classifying financial management customers, it is unable to adapt to the real-time update requirements of large-scale customer data, and the recommended products do not meet the needs, resulting in a decline in customer satisfaction.

[0158] Optionally, the data dimensionality reduction unit includes: a kernel matrix calculation module, which is used to calculate the kernel matrix of the preprocessed original data set using a selected kernel function; a centralization processing module, which is used to centralize the kernel matrix and perform eigendecomposition on the centralized kernel matrix to obtain a set of eigenvalues ​​and an eigenvector corresponding to each eigenvalue; a normalization processing module, which is used to normalize the eigenvectors; and an eigenvector selection module, which is used to sort all the normalized eigenvectors from large to small, select the eigenvectors corresponding to the first P eigenvalues, and obtain the reduced data point set, where P is a positive integer greater than 1.

[0159] Optionally, the outlier removal unit includes: a sample set acquisition module, which is used to use a local density factor algorithm to obtain a sample set that is closest to the data point to obtain a k-th nearest neighbor set, where k is a positive integer; obtain the k-th reachable distance between the data point and another data point; a local reachable density calculation module, which is used to calculate the local reachable density of key parameters of the associated data point based on the size of the k-th nearest neighbor set and the k-th reachable distance; and an outlier factor calculation module, which is used to calculate the outlier factor of the data point based on the local reachable density of the key parameters of the data point, the size of the k-th nearest neighbor set, and the average value of the local reachable density of another data point.

[0160] Optionally, the outlier removal unit includes: an outlier confirmation module, which is used to confirm that a data point is an outlier when the absolute value of the difference between the outlier factor and the preset outlier threshold is greater than the preset absolute value threshold, and to use the data value corresponding to the data point as the abnormal data value, and to remove the abnormal data value in the data point set after dimensionality reduction; a normal data point confirmation module, which is used to confirm that the data point belongs to a normal cluster when the absolute value of the difference between the outlier factor and the preset outlier threshold is less than or equal to the preset absolute value threshold, and to retain the data value corresponding to the data point.

[0161] Optionally, the cluster center determination unit includes: a directed graph construction module, which is used to use each data point in the data point set after dimensionality reduction as a node to construct a directed connection graph; a node connection strength calculation module, which is used to calculate the node connection strength based on the distance between each node in the directed connection graph and the control parameter, wherein the control parameter is a pre-set rate for adjusting the rate at which the connection strength between data points changes with distance; a linear equation group construction module, which is used to construct a linear equation group based on the calculated node connection strength, damping factor, transfer matrix generated by Markov chain traversal of nodes, and cluster reference PageRank value of each node transmitted through the link relationship.

[0162] Optionally, the cluster center determination unit also includes: an equation group iterative update module, which is used to iteratively update each vector element in the linear equation group using a preset iterative strategy; an update monitoring module, which is used to monitor the change of the linear equation group during the iterative update process until the convergence condition is met, and output the cluster reference PageRank value of each node, wherein the convergence condition includes: the element centroid change parameter is less than the preset parameter change threshold or the maximum number of iterative updates is reached; a cluster center selection module, which is used to sort the cluster reference PageRank values ​​of all nodes from large to small, and select the top F data points as the initial cluster centers, where F is a positive integer.

[0163] Optionally, the customer classification unit includes: an iterative parameter acquisition module for obtaining a pre-set centroid change threshold and a maximum number of clustering iterations; a data point attribution module for calculating the distance from each data point to each centroid in the data point set after dimensionality reduction, and assigning each data point to the cluster to which the nearest centroid belongs according to the distance; a new centroid selection module for recalculating the centroid of each cluster, and taking the mean of all data points in the current cluster as the new centroid; iteratively executing the data point attribution step and the new centroid selection step until the clustering convergence condition is reached, wherein the clustering convergence condition includes: the number of iterations reaches the maximum number of clustering iterations, the new centroid change value is less than the centroid change threshold, or the cluster assignment no longer changes; a customer classification module for determining the financial management customer classification result based on the cluster to which the centroid of each data point belongs determined by the last iterative execution.

[0164] Optionally, the customer classification device based on machine learning also includes: a missing value processing unit, which is used to perform missing value processing on the original data set after obtaining the original data set of financial management customers, wherein the missing value processing includes: mean filling, mode filling and deletion of invalid missing values; a numerical conversion unit, which is used to convert the text information in the original data set into numerical information; and a standardization unit, which is used to standardize the sample data in the original data set based on the standard deviation of the original data.

[0165] The above-mentioned customer group classification device based on machine learning can also include a processor and a memory. The above-mentioned data dimensionality reduction unit 41, outlier removal unit 42, cluster center point determination unit 43, customer group classification unit 44, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions.

[0166] The processor includes a core that retrieves the corresponding program unit from the memory. One or more cores can be configured, and the kernel parameters can be adjusted to classify financial management customer groups based on the kernel principal component analysis algorithm and the improved clustering algorithm.

[0167] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0168] Example 3

[0169] An embodiment of the present application may provide an electronic device, Figure 5 : is a structural block diagram of an electronic device that performs a customer group classification method based on machine learning according to an embodiment of the present application. Figure 5 As shown, the electronic device may include: one or more ( Figure 5Only one is shown) processor 502, memory 504, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0170] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the customer classification method and device based on machine learning in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned customer classification method based on machine learning. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, corporate intranet, local area network, mobile communication network and combinations thereof.

[0171] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: obtain the original data set of the financial management customer, and use the kernel principal component analysis algorithm to reduce the dimension of the original data set to obtain a data point set after dimensionality reduction, wherein the data point set after dimensionality reduction includes multiple data points and the data value corresponding to each data point; use the local density factor algorithm to calculate the outlier factor of each data point, and remove the abnormal data values ​​in the data point set after dimensionality reduction according to the outlier factor; calculate the connection strength between nodes according to the relationship between the data points, establish a linear equation group, and use a preset iterative strategy to solve the linear equation group until the convergence condition is met to obtain the initial cluster center point, wherein each initial cluster center point serves as the centroid of the clustering operation; perform cluster analysis based on the initial cluster center point, and classify each data point in the data point set after dimensionality reduction into the cluster to which the nearest centroid belongs until the cluster convergence condition is met, and obtain the classification result of the financial management customer group.

[0172] It can be understood by those skilled in the art that Figure 5 The structure shown is for illustration only, and the electronic device may also be a smart phone, a tablet computer, a PDA, a mobile internet device (MID), a PAD or other terminal device. Figure 5 It does not limit the structure of the above electronic device. For example, the electronic device may also include Figure 5 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 5 Different configurations shown.

[0173] Those skilled in the art will understand that all or part of the steps in the various machine learning-based customer classification methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0174] Example 4

[0175] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the customer group classification method based on machine learning provided in the first embodiment.

[0176] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the machine learning-based customer classification methods in the above-mentioned embodiment 1.

[0177] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0178] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the customer group classification method based on machine learning described in each embodiment of the present application.

[0179] The present application also provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the customer group classification method based on machine learning described in each embodiment of the present application are implemented.

[0180] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0181] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0182] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0183] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0184] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0185] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

[0186] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A customer group classification method based on machine learning, characterized in that: include: Obtaining an original data set of financial management customers, and performing dimensionality reduction processing on the original data set using a kernel principal component analysis algorithm to obtain a reduced-dimensionality data point set, wherein the reduced-dimensionality data point set includes multiple data points and a data value corresponding to each data point; Calculating the outlier factor of each data point using a local density factor algorithm, and removing abnormal data values ​​from the data point set after dimensionality reduction according to the outlier factor; Calculate the connection strength between nodes based on the relationship between data points, establish a linear equation system, and use a preset iterative strategy to solve the linear equation system until the convergence condition is met to obtain initial cluster center points, wherein each initial cluster center point serves as the centroid of the clustering operation; Cluster analysis is performed based on the initial cluster center point, and each data point in the dimensionality-reduced data point set is classified into a cluster to which the nearest centroid belongs until the cluster convergence condition is reached, thereby obtaining a classification result of the financial management customer group.

2. The customer group classification method based on machine learning according to claim 1, characterized in that: The step of performing dimensionality reduction processing on the original data set using a kernel principal component analysis algorithm to obtain a data point set after dimensionality reduction includes: Calculating a kernel matrix of the preprocessed original data set using a selected kernel function; Centralizing the kernel matrix and performing eigendecomposition on the centralized kernel matrix to obtain an eigenvalue set and an eigenvector corresponding to each eigenvalue; performing normalization processing on the feature vector; All the normalized eigenvectors are sorted from large to small, and the eigenvectors corresponding to the first P eigenvalues ​​are selected to obtain the data point set after dimensionality reduction, where P is a positive integer greater than 1.

3. The customer group classification method based on machine learning according to claim 1, characterized in that: The step of calculating the outlier factor of each data point using a local density factor algorithm comprises: A local density factor algorithm is used to obtain a sample set that is k-th closest to the data point to obtain a k-nearest neighbor set, where k is a positive integer; Obtaining the kth reachable distance between the data point and another data point; Calculating a local reachability density of a key parameter associated with the data point based on the size of the k nearest neighbor set and the kth reachable distance; The outlier factor of the data point is calculated based on the key parameters of the data point, the local reachable density, the size of the k-nearest neighbor set, and the average value of the local reachable density of another data point.

4. The customer group classification method based on machine learning according to claim 3, characterized in that: The step of removing abnormal data values ​​in the data point set after dimensionality reduction according to the outlier factor comprises: When the absolute value of the difference between the outlier factor and the preset outlier threshold is greater than the preset absolute value threshold, the data point is confirmed to be an outlier, the data value corresponding to the data point is used as an abnormal data value, and the abnormal data value is removed from the data point set after dimensionality reduction; When the absolute value of the difference between the outlier factor and the preset outlier threshold is less than or equal to the preset absolute value threshold, it is confirmed that the data point belongs to a normal cluster, and the data value corresponding to the data point is retained.

5. The customer group classification method based on machine learning according to claim 1, characterized in that: The steps of calculating the connection strength between nodes based on the relationship between data points and establishing a linear equation system include: Taking each data point in the dimensionality-reduced data point set as a node, and constructing a directed connection graph; Calculating the connection strength between nodes based on the distance between nodes in the directed connection graph and a control parameter, wherein the control parameter is a preset rate for adjusting the rate at which the connection strength between data points changes with distance; The linear equation group is constructed based on the calculated inter-node connection strength, damping factor, transfer matrix generated by Markov chain traversal of nodes, and cluster reference PageRank value of each node transmitted through link relationship.

6. The customer group classification method based on machine learning according to claim 5, characterized in that: The step of solving the linear equations by using a preset iterative strategy until a convergence condition is satisfied to obtain an initial cluster center point comprises: Iteratively updating each vector element in the linear equation system using a preset iterative strategy; Monitor the change of the linear equation system during the iterative update process until the convergence condition is met, and output the cluster reference PageRank value of each node, wherein the convergence condition includes: the element centroid change parameter is less than a preset parameter change threshold or the maximum number of iterative updates is reached; The cluster reference PageRank values ​​of all the nodes are sorted from large to small, and the top F data points are selected as the initial cluster centers, where F is a positive integer.

7. The customer group classification method based on machine learning according to claim 1, characterized in that: The step of classifying each data point in the dimensionality-reduced data point set into a cluster to which the nearest centroid belongs until a cluster convergence condition is reached to obtain a classification result of the financial management customer group includes: Obtain the pre-set centroid change threshold and maximum number of clustering iterations; The data point assignment step is to calculate the distance between each data point and each centroid in the data point set after dimensionality reduction, and assign each data point to the cluster to which the nearest centroid belongs according to the distance; In the new centroid selection step, the centroid of each cluster is recalculated and the mean of all data points in the current cluster is taken as the new centroid; Iteratively executing the data point attribution step and the new centroid selection step until a clustering convergence condition is reached, wherein the clustering convergence condition includes: the number of iterations reaches the maximum number of clustering iterations, the new centroid change value is less than the centroid change threshold, or the cluster assignment no longer changes; Based on the cluster to which the centroid of each data point belongs determined by the last iteration, the classification result of the financial management customer group is determined.

8. The customer group classification method based on machine learning according to claim 1, characterized in that: After obtaining the original data set of financial management customers, it also includes: Performing missing value processing on the original data set, wherein the missing value processing includes: filling the mean, filling the mode, and deleting invalid missing values; Converting the text information in the original data set into numerical information; Based on the standard deviation of the original data, the sample data in the original data set is standardized.

9. A customer group classification device based on machine learning, characterized in that: include: A data dimensionality reduction unit is used to obtain an original data set of financial management customers and perform dimensionality reduction processing on the original data set using a kernel principal component analysis algorithm to obtain a reduced-dimensionality data point set, wherein the reduced-dimensionality data point set includes multiple data points and a data value corresponding to each data point; an outlier removal unit, configured to calculate an outlier factor of each of the data points using a local density factor algorithm, and remove outlier data values ​​from the dimensionality-reduced data point set according to the outlier factor; A cluster center determination unit is used to calculate the connection strength between nodes based on the relationship between data points, establish a linear equation system, and solve the linear equation system using a preset iterative strategy until the convergence condition is met, thereby obtaining initial cluster centers, wherein each initial cluster center serves as the centroid of the clustering operation; The customer group classification unit is used to perform cluster analysis based on the initial cluster center point, classify each data point in the dimensionality-reduced data point set into the cluster to which the nearest centroid belongs until the cluster convergence condition is reached, and obtain the financial management customer group classification result.

10. An electronic device, characterized in that: The invention comprises one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the customer group classification method based on machine learning as described in any one of claims 1 to 8.

11. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the customer group classification method based on machine learning according to any one of claims 1 to 8 are implemented.

Images