Anchors-based clustering guidance data classification method, device and equipment
By constructing a joint iterative optimization of the anchor point matrix and the cluster center matrix, the problem of fuzzy C-means clustering technology being sensitive to initial conditions is solved, achieving high-quality data classification results and efficient data processing, which is suitable for large-scale datasets.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- EAST CHINA JIAOTONG UNIVERSITY
- Filing Date
- 2026-01-19
- Publication Date
- 2026-05-08
AI Technical Summary
Existing fuzzy C-means clustering techniques are sensitive to initial conditions and are prone to getting stuck in suboptimal solutions, resulting in unstable and inaccurate data classification results, making them difficult to apply to large-scale datasets.
By constructing an anchor point matrix and a cluster center matrix, a joint iterative optimization method is adopted, and the anchor points and cluster centers are updated using the gradient descent algorithm. A specific distance metric is introduced during the optimization process to reduce the sensitivity to initial conditions and improve the robustness and consistency of the classification results.
It achieves high-quality data classification results, reduces the cost of manual annotation, improves the transparency and trustworthiness of data processing, enhances the classification accuracy and applicability of complex data, and shortens the computation time.
Smart Images
Figure CN121542881B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of electronic digital data processing technology, and in particular to a data classification method, apparatus and equipment based on anchor-guided clustering. Background Technology
[0002] In numerous technological scenarios involving the automated analysis and processing of unlabeled data, such as image content recognition, industrial sensor data classification, or biometric classification, clustering analysis is a crucial data processing technique. Among these, fuzzy C-means clustering, as a solution capable of outputting the probability of a sample belonging to each category, has been widely applied in these scenarios due to its ability to handle data with unclear category boundaries. Fuzzy C-means clustering is a distance-based soft clustering algorithm that uses fuzzy membership degrees to assign data points to multiple clusters, representing the probability of a point belonging to each cluster, ultimately achieving data classification output.
[0003] However, in practical applications, it has been found that the quality of the final classification results produced by using existing fuzzy C-means clustering techniques for data processing is unstable. Specifically, the data processing is extremely sensitive to the initial conditions of computational parameters such as cluster centers; different initial settings can lead to significantly different classification results. Furthermore, it is highly sensitive to noise and outliers, affecting the stability of data classification. In addition, during the optimization process, the existing data processing flow can cause the model to converge to a suboptimal local minimum, and its solution trajectory is prone to premature simplification, limiting its ability to obtain more accurate classification results on complex data.
[0004] The existing technology suffers from the technical problem that the data processing flow is sensitive to initial conditions and is prone to falling into suboptimal solutions, resulting in inaccurate and unstable final data classification results. Summary of the Invention
[0005] The purpose of this application is to provide a data classification method, apparatus, and device based on anchor-guided clustering. By constructing and jointly optimizing the data matrix, anchor matrix, and cluster center matrix, the optimization variables are transformed from a large membership matrix into a much smaller anchor and center matrix. This effectively solves the technical problems of existing fuzzy clustering methods, which are sensitive to initial conditions and prone to falling into suboptimal solutions, resulting in unstable and inaccurate classification results, and cannot be directly solved using gradient descent algorithms, making them difficult to apply to large-scale datasets. This achieves enhanced classification accuracy for complex data and applicability to different data scale scenarios.
[0006] To achieve the above objectives, this application provides the following solution:
[0007] In a first aspect, this application provides a data classification method based on anchor-guided clustering, comprising: acquiring an original dataset to be classified, and converting each sample data in the original dataset into a numerical vector; arranging the numerical vectors of all samples to construct a data matrix; initializing a cluster center matrix according to a preset number of cluster categories, initializing an anchor matrix according to a preset number of anchor points, and setting fuzzy coefficients; performing a preset joint iterative optimization based on the data matrix, the anchor matrix, and the cluster center matrix to obtain an optimized anchor matrix and an optimized cluster center matrix; calculating the fuzzy membership degree of each sample vector in the data matrix to each cluster center according to the optimized cluster center matrix; determining the category with the largest fuzzy membership degree value corresponding to each sample vector in the data matrix, and using it as the sample. The final category label of the vector is output as the classification result for all sample vectors. The preset joint iterative optimization includes multiple iterations, iteratively executing the following steps until convergence: calculating a first distance metric between the data matrix and the anchor matrix, a second distance metric between the anchor matrix and the cluster center matrix, and a third distance metric between the data matrix and the cluster center matrix; determining the gradient of the anchor matrix based on the first and second distance metrics; determining the gradient of the cluster center matrix based on the second and third distance metrics; updating the values stored in the anchor matrix based on the gradient of the anchor matrix; updating the values stored in the cluster center matrix based on the gradient of the cluster center matrix to obtain the optimized cluster center matrix.
[0008] Optionally, the calculation of the first distance metric, the second distance metric, and the third distance metric together constitute a joint optimization objective; wherein the third distance metric constitutes the dominant term of the joint optimization objective, and the first distance metric and the second distance metric constitute auxiliary terms used to guide the optimization process.
[0009] Optionally, the fuzzy membership degree The calculation formula is: Where c represents the number of clusters, r represents the fuzzy coefficient, and m represents the number of anchor points. This represents the i-th sample vector in the data matrix X. It is represented as the j-th cluster center vector in the cluster center matrix M. It is represented as the k-th cluster center vector in the cluster center matrix M.
[0010] Optionally, the first distance metric, the second distance metric, and the third distance metric are optimized using a joint optimization objective function L; the joint optimization objective function L is:
[0011]
[0012] Where Z represents the anchor point matrix, It is represented as the j-th anchor vector in the anchor matrix Z, where n is the number of samples.
[0013] Optionally, determining the gradient of the anchor matrix and updating the values stored in the anchor matrix, as well as determining the gradient of the cluster center matrix and updating the values stored in the cluster center matrix, are all implemented using gradient descent; wherein, the gradient of the anchor matrix... and the gradient of the cluster center matrix All are calculated based on the joint optimization objective function L;
[0014] Anchor points in the anchor point matrix The update formula is: ;
[0015] The cluster center matrix contains cluster centers. The update formula is: ;
[0016] in, Let t represent the learning rate, and t be the t-th iteration.
[0017] Optionally, the fuzzy coefficient may be greater than 1.
[0018] Secondly, this application provides a data classification device based on anchor-guided clustering, comprising: a data acquisition and construction module configured to acquire an original dataset to be classified, convert each sample data in the original dataset into a numerical vector, and arrange the numerical vectors of all samples to construct a data matrix; a parameter initialization module configured to initialize a cluster center matrix according to a preset number of cluster categories, initialize an anchor matrix according to a preset number of anchor points, and set fuzzy coefficients; a joint iterative optimization module configured to perform a preset joint iterative optimization based on the data matrix, the anchor matrix, and the cluster center matrix to obtain an optimized anchor matrix and an optimized cluster center matrix; a fuzzy membership calculation module configured to calculate the fuzzy membership degree of each sample vector in the data matrix to each cluster center based on the optimized cluster center matrix; and a classification decision and output module configured to determine the fuzzy membership degree of each sample vector in the data matrix. The category with the largest attribute value is selected as the final category label for the sample vector, and the final category label of all sample vectors is output as the classification result. The joint iterative optimization module is also configured to perform multiple iterations until convergence. The joint iterative optimization module includes: a distance calculation unit configured to calculate a first distance metric between the data matrix and the anchor matrix, a second distance metric between the anchor matrix and the cluster center matrix, and a third distance metric between the data matrix and the cluster center matrix; a gradient determination unit configured to determine the gradient of the anchor matrix based on the first and second distance metrics; and determine the gradient of the cluster center matrix based on the second and third distance metrics; and a value update unit configured to update the values stored in the anchor matrix based on the gradient of the anchor matrix; and update the values stored in the cluster center matrix based on the gradient of the cluster center matrix to obtain the optimized cluster center matrix.
[0019] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the data classification method based on anchor-guided clustering as described above.
[0020] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the data classification method based on anchor-guided clustering described above.
[0021] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the data classification method based on anchor-guided clustering described above.
[0022] According to the specific embodiments provided in this application, the following technical effects are disclosed:
[0023] This application provides a data classification method, apparatus, and device based on anchor-guided clustering. It can automatically complete the entire process from raw unstructured data, including vectorization, matrix construction, model optimization, and final label output. The entire process requires no manual data labeling and directly outputs high-quality classification results, significantly reducing the cost of manual labeling and the analytical threshold. Furthermore, the results of traditional clustering methods often fluctuate due to different initializations. This application significantly enhances the robustness of the output results by introducing an anchor-guided joint optimization mechanism. Through specific distance metric calculations and collaborative optimization, it can more effectively capture the inherent structure and relationships of the data. In addition, the final output of this data classification method not only includes hard classification labels but also provides fuzzy membership degrees for each sample belonging to each category, providing a confidence reference. Users can assess the reliability of the classification based on the membership degrees, enhancing the transparency and trustworthiness of the entire data processing process. Furthermore, by introducing an anchor-guided collaborative update mechanism, the sensitivity of the data processing flow to the initial settings is significantly reduced. Even when starting from different random starting points, the final classification results show higher consistency. Moreover, the optimization process transforms the optimization variables from a large membership matrix into a much smaller anchor and center matrix. This directly reduces the amount of memory data that needs to be accessed and updated in each iteration, reduces the computational load, shortens the total computation time required to classify the entire dataset, and improves the data processing throughput. Attached Figure Description
[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0025] Figure 1 This is an application environment diagram of a data classification method based on anchor-guided clustering in one embodiment of this application;
[0026] Figure 2 A flowchart illustrating a data classification method based on anchor-guided clustering, provided as an embodiment of this application;
[0027] Figure 3 for Figure 2 A detailed flowchart of step 15;
[0028] Figure 4 This is a schematic diagram illustrating the principle of anchor-guided clustering optimization in one embodiment of this application;
[0029] Figure 5 A schematic diagram of the functional modules of a data classification device based on anchor-guided clustering provided in an embodiment of this application;
[0030] Figure 6 for Figure 5 A detailed functional module diagram of the joint iterative optimization module;
[0031] Figure 7 This application provides a schematic diagram of the structure of a computer device according to one embodiment. Detailed Implementation
[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0033] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0034] This application provides a data classification method based on anchor-guided clustering, which can be applied to, for example... Figure 1In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send the raw dataset to be classified to server 102. Server 102 receives the raw dataset, converts each sample data in the raw dataset into a numerical vector; arranges all the numerical vectors of the samples to construct a data matrix; initializes the cluster center matrix according to the preset number of cluster categories, initializes the anchor matrix according to the preset number of anchor points, and sets fuzzy coefficients; based on the data matrix, anchor matrix, and cluster center matrix, performs a preset joint iterative optimization to obtain the optimized anchor matrix and optimized cluster center matrix; according to the optimized cluster center matrix, calculates the fuzzy membership degree of each sample vector in the data matrix belonging to each cluster center; determines the category with the largest fuzzy membership degree value for each sample vector in the data matrix, and uses it as the final category label for the sample vector; outputs the final category label of all sample vectors as the classification result. Server 102 can feed back the classification results to terminal 101. In addition, in some embodiments, the classification results can also be implemented by server 102 or terminal 101 alone. For example, terminal 101 can directly process the original dataset to be processed, or server 102 can obtain the original dataset to be classified from the data storage system.
[0035] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.
[0036] In one exemplary embodiment, such as Figure 2 As shown, a data classification method based on anchor-guided clustering is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 102 as an example, a data classification method based on anchor-guided clustering includes the following steps:
[0037] S11, Obtain the original dataset to be classified;
[0038] S12 converts each sample data in the original dataset into a numerical vector;
[0039] S13, arrange the numerical vectors of all samples to construct a data matrix;
[0040] S14. Initialize the cluster center matrix according to the preset number of cluster categories, initialize the anchor point matrix according to the preset number of anchor points, and set the fuzzy coefficient.
[0041] S15, based on the data matrix, anchor matrix and cluster center matrix, performs a preset joint iterative optimization to obtain the optimized anchor matrix and the optimized cluster center matrix;
[0042] S16, Based on the optimized cluster center matrix, calculate the fuzzy membership degree of each sample vector in the data matrix to each cluster center;
[0043] S17: Determine the category with the largest fuzzy membership value for each sample vector in the data matrix, and use it as the final category label for the sample vector. Output the final category labels for all sample vectors as the classification result.
[0044] Among them, such as Figure 3 As shown, the pre-defined joint iterative optimization in step 15 includes multiple iterations, iteratively executing the following steps until convergence:
[0045] S151, calculate the first distance metric between the data matrix and the anchor matrix, the second distance metric between the anchor matrix and the cluster center matrix, and the third distance metric between the data matrix and the cluster center matrix;
[0046] S152, determine the gradient of the anchor matrix based on the first and second distance metrics; determine the gradient of the cluster center matrix based on the second and third distance metrics;
[0047] S153, based on the gradient of the anchor matrix, update the values stored in the anchor matrix; based on the gradient of the cluster center matrix, update the values stored in the cluster center matrix to obtain the optimized cluster center matrix.
[0048] Implementing steps 11 to 17 above automatically completes the entire process from raw unstructured data, including vectorization, matrix construction, model optimization, and final label output. The entire process requires no manual data labeling and directly outputs high-quality classification results, significantly reducing the cost of manual labeling and the analytical threshold. Furthermore, the results of traditional clustering methods often fluctuate due to different initializations. This application significantly enhances the robustness of the output results by introducing an anchor-guided joint optimization mechanism. Through specific distance metric calculations and collaborative optimization, it can more effectively capture the inherent structure and relationships of the data. In addition, the final output of this data classification method not only includes hard classification labels but also provides fuzzy membership degrees for each sample belonging to each category, providing a confidence reference. Users can assess the reliability of the classification based on the membership degrees, enhancing the transparency and trustworthiness of the entire data processing process. Furthermore, by implementing the joint iterative optimization process of steps S151 to S153, and by introducing an anchor-guided collaborative update mechanism, the sensitivity of the data processing flow to the initial settings is significantly reduced. Even when starting from different random starting points, the final output classification results show higher consistency. Moreover, the optimization process transforms the optimization variables from a large membership matrix into a much smaller anchor and center matrix, which directly reduces the amount of memory data that needs to be accessed and updated in each iteration, thus reducing the computational load. At the same time, the gradient descent algorithm is used for solving the problem, which is suitable for processing large-scale datasets. This shortens the total computation time required to classify the entire dataset and significantly improves the data processing throughput.
[0049] In specific implementation, step 11 involves the original dataset to be classified. This original dataset can originate from various practical application scenarios, such as: image data (a collection of two-dimensional RGB or grayscale images acquired through cameras or scanners); industrial time-series data (time-series data such as equipment vibration signals, temperature sequences, and pressure waveforms collected from sensor networks); text data (a collection of text extracted from documents, web pages, or logs); and bioinformatics data (gene expression profiles, protein sequence data, etc.).
[0050] In practice, step 12 involves feature extraction and numerical transformation of each original sample data, representing the original sample data as a d-dimensional numerical vector. , where i is the sample index. For example, if it is two-dimensional image data acquired by a camera, the two-dimensional image data needs to be converted into a one-dimensional vector for processing first. If the data itself is one-dimensional, no conversion is required.
[0051] In the specific implementation, in step 13, the numerical vectors of all n samples are... Arrange the data rows to construct a regular data matrix in memory. It is typically an n×d matrix. Specifically, the data matrix... , where n is the number of data samples and d is the data dimension of the samples.
[0052] In specific implementation, in step 14, the number of cluster categories *c* is a positive integer representing the desired number of categories into which the data is divided; for example, in image classification, *c* might correspond to the number of object types; in customer segmentation, *c* corresponds to the number of customer groups. The number of anchor points *m* is a positive integer representing the number of intermediate guiding points introduced, used to balance model complexity and guiding capability. The fuzziness coefficient *r* is a real number greater than 1, used to control the degree of fuzziness in the clustering results; the closer *r* is to 1, the more the clustering results tend towards hard partitioning; the larger *r* is, the more fuzzy the membership. Based on the preset number of cluster categories *c* and data dimension *d*, a c×d cluster center matrix *M* is created and assigned values as the initial iteration values of the cluster center matrix *M*. Based on the preset number of anchor points *m* and data dimension *d*, an m×d anchor matrix *Z* is created and assigned values as the initial iteration values of the anchor matrix *Z*. The initialization strategy for the anchor matrix *Z* can be similar to that of the cluster center matrix *M*, both using a random initialization method.
[0053] In specific implementation, step 15 involves a joint iterative optimization process based on a specific joint optimization objective function L. The joint optimization objective function L consists of three parts, corresponding to the aggregation of the first, second, and third distance metrics calculated in step S151. The expression for the joint optimization objective function L is:
[0054]
[0055] in, This represents the i-th sample vector in the data matrix X. It is represented as the j-th cluster center vector in the cluster center matrix M. Let be the specific k-th cluster center vector in the cluster center matrix M, where the first term reflects the overall degree of difference between all data points and all anchor points; the second term reflects the overall degree of difference between all anchor points and all cluster centers; and the third term reflects the overall degree of difference between all data points and all cluster centers.
[0056] The third term acts as the dominant term, its minimization directly corresponding to seeking the configuration that makes the data points as close as possible to the final cluster centers, which constitutes the core objective of the clustering task. The first and second terms serve as guiding terms, establishing a guiding structure between the data points and cluster centers by introducing and optimizing the anchor matrix Z. In the optimization process, the guiding terms play two roles: first, they help explore the intrinsic distribution of the data, providing richer structural information for the optimization of cluster centers; second, through the intermediate variable of the anchor, they effectively broaden the optimization path, reducing the risk of the optimization process getting trapped in poor local optima, thus guiding the optimization of the dominant term to a better solution space.
[0057] Furthermore, in step S152, after obtaining the first distance metric, the second distance metric, and the third distance metric, the anchor point matrix Z and the cluster center matrix M are jointly iteratively updated.
[0058] For the k-th anchor point in the anchor point matrix Z The gradient of the anchor matrix Z The calculation formula is:
[0059]
[0060] For the k-th cluster center in the cluster center matrix M The gradient of the cluster center matrix M The calculation formula is:
[0061]
[0062] Furthermore, in step S153, after calculating the gradients of all anchor points and cluster centers, the values stored in the two matrices are updated synchronously. The update follows the gradient descent principle, that is, it proceeds in the opposite direction of the gradient with a step size of [missing value]. Adjust the (learning rate).
[0063] Anchor point The update formula is: ;
[0064] Cluster Center The update formula is: ;
[0065] Where t is the t-th iteration, and the learning rate η is a preset positive hyperparameter used to control the magnitude of each update. The value of the learning rate can be selected based on experience.
[0066] In specific implementation, step 16 involves calculating the fuzzy membership degree of each sample i to each category j. The calculation formula is: It transforms abstract cluster centers into specific, quantified class assignment probabilities for each sample.
[0067] In specific implementation, in step 17, after obtaining the fuzzy membership degree of each sample i to each category j... Then, the final category of each sample is determined, and the decision rule is determined to be the maximum membership principle, which is mathematically expressed as: , This represents the value of the label matrix F at (i, j). If =1, indicating the i-th data point. The category it belongs to is class j. The final output category label is a clear, concise, and directly applicable technical result, such as for evaluating classification accuracy, data archiving, or driving subsequent decision-making systems, thereby achieving the technical goal of intelligent and automated classification of the original dataset.
[0068] In one exemplary embodiment, such as Figure 4 As shown, a schematic diagram of the anchor-guided clustering optimization principle is provided. First... Figure 4 The algorithm learns from multiple original data points on the far left to generate a smaller set of anchor points. These anchor points are then used to capture the inherent substructure of the data. An optimization process guided by these anchor points is then implemented to prevent data points from getting trapped in local optima. Finally, the data points are assigned to their respective clusters based on their affiliation with the optimized cluster centers. Thus, Figure 4 It visually demonstrates the technical concept of improving clustering performance by introducing anchor points as intermediate guiding structures.
[0069] In one exemplary embodiment, such as Figure 5 As shown, a data classification device based on anchor-guided clustering is provided, comprising:
[0070] The data acquisition and construction module 21 is configured to acquire the original dataset to be classified, convert each sample data in the original dataset into a numerical vector, and arrange the numerical vectors of all samples to construct a data matrix.
[0071] The parameter initialization module 22 is configured to initialize the cluster center matrix according to the preset number of cluster categories, initialize the anchor point matrix according to the preset number of anchor points, and set the fuzzy coefficients.
[0072] The joint iterative optimization module 23 is configured to perform a preset joint iterative optimization based on the data matrix, the anchor matrix, and the cluster center matrix to obtain the optimized anchor matrix and the optimized cluster center matrix.
[0073] The fuzzy membership calculation module 24 is configured to calculate the fuzzy membership degree of each sample vector in the data matrix to each cluster center based on the optimized cluster center matrix.
[0074] The classification decision and output module 25 is configured to determine the category with the largest fuzzy membership value for each sample vector in the data matrix, and use it as the final category label for the sample vector, and output the final category label for all sample vectors as the classification result.
[0075] Among them, such as Figure 6 As shown, the joint iterative optimization module 23 includes:
[0076] The distance calculation unit 231 is configured to perform calculations of a first distance metric between the data matrix and the anchor matrix, a second distance metric between the anchor matrix and the cluster center matrix, and a third distance metric between the data matrix and the cluster center matrix;
[0077] The gradient determination unit 232 is configured to perform the following operations: determine the gradient of the anchor matrix based on the first distance metric and the second distance metric; and determine the gradient of the cluster center matrix based on the second distance metric and the third distance metric.
[0078] The numerical update unit 233 is configured to perform the following operations: update the numerical values stored in the anchor matrix according to the gradient of the anchor matrix; update the numerical values stored in the cluster center matrix according to the gradient of the cluster center matrix, and obtain the optimized cluster center matrix.
[0079] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The computer device's database stores the raw dataset to be classified. The computer device's I / O interfaces are used for exchanging information between the processor and external devices. The computer device's communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a data classification method based on anchor-guided clustering.
[0080] Those skilled in the art will understand that Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0081] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0082] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0083] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0084] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0085] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Furthermore, any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory.
[0086] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0087] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A data classification method based on anchor-guided clustering, characterized in that, include: Obtain the original dataset to be classified, and convert each sample data in the original dataset into a numerical vector; Arrange the numerical vectors of all samples to construct a data matrix; Initialize the cluster center matrix according to the preset number of cluster categories, initialize the anchor point matrix according to the preset number of anchor points, and set the fuzzy coefficients; Based on the data matrix, the anchor matrix, and the cluster center matrix, a preset joint iterative optimization is performed to obtain the optimized anchor matrix and the optimized cluster center matrix. Based on the optimized cluster center matrix, calculate the fuzzy membership degree of each sample vector in the data matrix to each cluster center; The class with the largest fuzzy membership value for each sample vector in the data matrix is determined and used as the final class label for the sample vector. The final class labels for all sample vectors are then output as the classification result. The pre-defined joint iterative optimization includes multiple iterations, iteratively executing the following steps until convergence: Calculate a first distance metric between the data matrix and the anchor matrix, a second distance metric between the anchor matrix and the cluster center matrix, and a third distance metric between the data matrix and the cluster center matrix; The gradient of the anchor point matrix is determined based on the first distance metric and the second distance metric; the gradient of the cluster center matrix is determined based on the second distance metric and the third distance metric. Update the values stored in the anchor matrix according to the gradient of the anchor matrix; update the values stored in the cluster center matrix according to the gradient of the cluster center matrix to obtain the optimized cluster center matrix. The calculation of the first distance metric, the second distance metric, and the third distance metric together constitutes a joint optimization objective; wherein, the third distance metric constitutes the dominant term of the joint optimization objective, and the first distance metric and the second distance metric constitute auxiliary terms used to guide the optimization process; The fuzzy membership degree The calculation formula is: Where c represents the number of clusters, and r represents the fuzzy coefficient. This represents the i-th sample vector in the data matrix X. and Each represents a cluster center in the cluster center matrix M, where j and k are index variables; The first distance metric, the second distance metric, and the third distance metric are optimized using a joint optimization objective function L; the joint optimization objective function L is: Where Z represents the anchor point matrix, It is represented as the j-th anchor vector in the anchor matrix Z, where n is the number of samples; The determination of the gradient of the anchor point matrix and the updating of the values stored in the anchor point matrix, as well as the determination of the gradient of the cluster center matrix and the updating of the values stored in the cluster center matrix, are all implemented using the gradient descent method; wherein, the gradient of the anchor point matrix... and the gradient of the cluster center matrix All are calculated based on the joint optimization objective function L; Anchor points in the anchor point matrix The update formula is: ; The cluster center matrix contains cluster centers. The update formula is: ; in, Let t represent the learning rate, and t be the t-th iteration.
2. The data classification method based on anchor-guided clustering according to claim 1, characterized in that, The fuzzy coefficient is greater than 1.
3. An apparatus for performing the anchor-guided clustering-based data classification method as described in claim 1 or 2, characterized in that, include: The data acquisition and construction module is configured to acquire the original dataset to be classified, convert each sample data in the original dataset into a numerical vector, and arrange the numerical vectors of all samples to construct a data matrix; The parameter initialization module is configured to initialize the cluster center matrix according to the preset number of cluster categories, initialize the anchor point matrix according to the preset number of anchor points, and set the fuzzy coefficients. The joint iterative optimization module is configured to perform a preset joint iterative optimization based on the data matrix, the anchor matrix, and the cluster center matrix to obtain the optimized anchor matrix and the optimized cluster center matrix. The fuzzy membership calculation module is configured to calculate the fuzzy membership degree of each sample vector in the data matrix to each cluster center based on the optimized cluster center matrix. The classification decision and output module is configured to determine the category with the largest fuzzy membership value for each sample vector in the data matrix, and use it as the final category label for the sample vector, and output the final category label for all sample vectors as the classification result. The joint iterative optimization module is also configured to perform multiple iterations until convergence; The joint iterative optimization module includes: The distance calculation unit is configured to perform calculations of a first distance metric between the data matrix and the anchor matrix, a second distance metric between the anchor matrix and the cluster center matrix, and a third distance metric between the data matrix and the cluster center matrix; The gradient determination unit is configured to perform the following actions: determining the gradient of the anchor matrix based on the first distance metric and the second distance metric; and determining the gradient of the cluster center matrix based on the second distance metric and the third distance metric. The numerical update unit is configured to perform the following operations: update the numerical values stored in the anchor matrix according to the gradient of the anchor matrix; update the numerical values stored in the cluster center matrix according to the gradient of the cluster center matrix, and obtain the optimized cluster center matrix.
4. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the data classification method based on anchor-guided clustering as described in any one of claims 1-2.
5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the data classification method based on anchor-guided clustering as described in any one of claims 1-2.
6. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the data classification method based on anchor-guided clustering as described in any one of claims 1-2.