Three-branch clustering method based on granules and balls

Through the three-branch clustering method based on particle ball, the data weight is adjusted using the particle ball model and the nearest neighbor relationship, and the problem of high-dimensional data consumption in the existing technology is solved, which improves clustering efficiency and reduces the sample Object allocation risks improve the quality of market segmentation results.

CN118917865APending Publication Date: 2024-11-08JIANGSU UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410703695.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-03
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the era of Internet big data, the existing clustering methods of two-branch decisions consume a lot of computing resources and time in high-dimensional data processing, and there is a risk that sample objects are forced to be allocated into a type of cluster, making it difficult to effectively process massive data.

Method used

Using the three-branch clustering method based on the particle ball, the particle ball model is constructed after preliminary clustering through the k-means algorithm, and the core domain and boundary domain are divided using the core and radius of the particle ball, the data weight is adjusted according to the nearest neighbor relationship, and the ball iteratively updates iteratively Center and radius, reducing unnecessary calculations.

Benefits of technology

It improves the efficiency of high-dimensional data clustering, reduces the risk of forced allocation of sample objects to a type of cluster, and improves the quality of market segmentation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917865A_ABST
    Figure CN118917865A_ABST
Patent Text Reader

Abstract

The invention discloses a three-branch clustering method based on granules and balls. The three-branch clustering method comprises the following steps: (1) acquiring information in a required market subdivision data set Pricerunner by utilizing Matlab; wherein the file format is mat; (2) analyzing consumer data information by using a k-means algorithm and clustering the consumer data information; (3) according to the obtained clustering result, constructing different consumer market granular ball models by using a three-branch clustering method, and obtaining a market segmentation result; according to the method, the unnecessary calculation amount is reduced to a certain extent, and meanwhile, the method can be adaptive to division of different data set regions in the clustered sub-markets, so that various performances of a market subdivision result are improved, the separation degree between the sub-markets is improved, and the internal structure between consumer data can be better revealed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of market segmentation, and in particular to a three-branch clustering method based on particle spheres. Background Art

[0002] Market segmentation is an important concept in marketing. It refers to dividing the entire market into different market segments according to certain specific standards or characteristics in order to better meet the needs of different consumer groups. Cluster analysis, as a data mining technology, can help marketers better understand market segmentation, identify potential consumer groups, and develop targeted marketing strategies. Based on this, domestic and foreign scholars have made many contributions: Prior art [1] (Qiu Huawei, et al. "Research on the Segmentation of Traditional Chinese Medicine Market in my country Based on Principal Components and Clustering Methods." Price Theory and Practice (2022).), Qiu Huawei et al. used the squared Euclidean distance and the inter-group connection measurement interval to perform hierarchical clustering on the sample data; Prior art [2] ( Bach M,Pivar J, B.Churn management intelecommunications:Hybrid approach using cluster analysis and decision trees[J].Journal of risk and financial management, 2021,14(11):544.), Bach M et al. combined k-means clustering with the CHAID decision tree algorithm to construct a classification model to identify the churn determinants in the clusters with the highest churn levels. The classification of relevant data made by the above two algorithms belongs to the scope of two-branch decision. There are clear boundaries between the clusters where the samples are located, and there are no objects that belong to both one cluster and another different cluster. At present, the era of Internet big data has arrived. It is risky to maintain the clustering of two-branch decisions and force the objects in the samples to be divided into one cluster. In addition, in the case of high-dimensional data, facing the various information brought by massive data, cluster analysis requires a lot of computing resources and time to process large-scale data sets. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a three-branch clustering method based on granular spheres, which can effectively reduce the large amount of calculations in the clustering process and reduce the risk of forcibly assigning sample objects to a cluster under uncertain information conditions, thereby improving the quality of market segmentation results.

[0004] Technical solution: The three-branch clustering method based on particle spheres described in the present invention comprises the following steps:

[0005] (1) Use Matlab to obtain the information in the required market segmentation data set Pricerunner; the file format is mat;

[0006] (2) Use the k-means algorithm to analyze and cluster consumer data information;

[0007] (3) Based on the clustering results obtained, the three-branch clustering method is used to construct different consumer market granular models to obtain market segmentation results.

[0008] Furthermore, in step (1), the information in the Pricerunner dataset includes: number of samples 35311, number of features 7, and number of categories 10.

[0009] Furthermore, step (2) is as follows: using the k-means algorithm, the consumer information dataset U = {x1, x2, ..., x 35311} Perform a preliminary clustering with an iteration number of 1 to obtain the preliminary market segmentation result C = {C1, C2, ..., C 10}; In the clustering results, any two parts of the cluster C i With C j does not contain the same data set, and any cluster is not an empty set.

[0010] Furthermore, step (3) includes the following steps:

[0011] (31) The center of the sphere is: The radius of the sphere is: r i = max{dist(x,c i )|x∈C i}, construct the initial consumer sub-market sphere model; the center of the sphere is the cluster C i The average value of the consumer data set in the cluster, i = 1, ..., 10; the radius of the sphere is the distance from the farthest data set in the cluster to the center of the sphere; |C i | is the current ball C i The number of consumer samples in, x is the current particle C i Consumer data samples in, dist(.) represents the vector x and c i The Euclidean distance between them, x represents the current consumer sub-market particle C i Consumer data sample in, c i Represented as a sphere C i The center of the sphere;

[0012] (32) Set the number of algorithm iterations n, the weight of the data set in the core domain of the sphere w, and the weight of the data set in the boundary domain 1-w, where 0 <w<1;

[0013] (33) Calculate the distance between the center of the sphere in the sphere model of the i-th initial consumer sub-market and the spheres of other sub-markets;

[0014] (34) Determine the neighboring spheres;

[0015] (35) Determine the stable domain and active domain;

[0016] (36) Update the sphere C for the stable domain i The center of the sphere c i ' and radius r i ', the formula is as follows:

[0017] r i '=max{dist(x,c i ')|x∈Co(C i )∪Fr(C i )'}, where |Co(C i )| represents the particle C i The number of consumer datasets in the core domain, |Fr(C i )'| represents the number of data sets in the new boundary domain after the consumer data sets in the original boundary domain are redistributed to the nearest neighbor, w represents the weight of the data sets in the domain, and x represents the consumer data sets contained in the domain;

[0018] The sample data in the remaining areas constitute the boundary domain and are allocated to the remaining sub-market spheres based on the principle of proximity.

[0019] (37) Repeat steps (33) to (36) k times and one iteration is completed. If the centers of all spheres c i If there is no change or the number of iterations reaches the preset threshold n, then extract Get the ball C i The core domain of Get the ball C i The final market segmentation result is: C = {(Co(C1), Fr(C1), (Co(C2), Fr(C2)…, (Co(C k ),Fr(C k )}.

[0020] Furthermore, step (34) is specifically as follows: Assume that the radius of the submarket particle sphere model C is r and the center of the sphere is c; if C i is the neighboring sphere of sphere C, then Among them, r = max{dist(x,c)|x∈C}, x is the consumer sample in the current cluster. Then Make Assume that x is on line segment cc i On, we have ||xc||+||xc i ||=||cc i ||; That is, ||xc||>||xc i ||; That is, if C i is the neighboring sphere of sphere C, then the consumer data in sphere C is adjusted to sphere C i The possibility of

[0021] If C j is not a neighboring sphere of sphere C, then In the submarket sphere model C, ||cx||≤r, ||cx||≤||xc j ||;

[0022] Furthermore, step (35) is as follows: Given any ball C i , the set of its neighboring sphere centers is {N Ci};like Then for any sphere C in the sphere center set j , center c j ∈{N Ci}, with c i is the center of the ball, is the stable domain covered by the radius, and the particle C i The area outside the stable domain is defined as the active domain; if Then the whole sphere C i All are stable domains.

[0023] Furthermore, in step (36), the proximity principle is as follows: i In the above example, after removing the data in the stable domain, the remaining data constitute the boundary domain of the sphere; the distances from the data in the boundary domain of the sphere to the centers of the 10 spheres are calculated in sequence, and the data are redistributed to the sphere model closest to it.

[0024] The three-branch clustering system based on particle spheres described in the present invention comprises:

[0025] Acquisition module: used to use Matlab to obtain the information in the required market segmentation data set Pricerunner; the file format is mat;

[0026] Clustering module: used to analyze and cluster consumer data using the k-means algorithm;

[0027] Three-branch clustering module: used to construct different consumer market granular models based on the obtained clustering results using the three-branch clustering method to obtain market segmentation results.

[0028] An electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, any one of the particle-sphere-based three-branch clustering methods described above is implemented.

[0029] A storage medium described in the present invention stores a computer program, and when the computer program is executed by a processor, it implements any one of the three-branch clustering methods based on particle spheres.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: by establishing a sphere model for each consumer sub-market, neighboring spheres are found for each sphere in an adaptive manner. The sphere is divided into a core domain and a boundary domain according to the situation of the neighboring spheres, and the consumer data in the core domain and the boundary domain are updated and iterated by assigning different weights. When the set threshold is reached or the center of the sphere no longer changes, the final market segmentation result is obtained. The consumer sub-market model constructed by this method can make appropriate adjustments to different types of consumer data sets during the clustering process, and flexibly modify the weights between consumer data sets in different regions; it can also improve the efficiency of clustering big data and high-dimensional data, and reduce some unnecessary calculations. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flow chart of the present invention;

[0032] Figure 2 This is a preliminary market segmentation flow chart of the present invention.

[0033] Figure 3 This is a flow chart of dividing the current particle sphere boundary domain and core domain of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0035] like Figure 1 As shown, the embodiment of the present invention provides a three-branch clustering method based on particle spheres, comprising the following steps:

[0036] (1) Use Matlab to obtain the information in the required market segmentation dataset Pricerunner; the file format is mat; the information in the dataset Pricerunner includes: number of samples 35311, number of features 7, number of categories 10

[0037] (2) Use the k-means algorithm to analyze and cluster consumer data information; specifically: Use the k-means algorithm to cluster the consumer information data set U = {x1, x2, ..., x 35311} Perform a preliminary clustering with an iteration number of 1 to obtain the preliminary market segmentation result C = {C1, C2, ..., C 10}; In the clustering results, any two parts of the cluster C i With C j does not contain the same data set, and any cluster is not an empty set. The specific process is as follows:

[0038] (21) Initialization: The algorithm first randomly selects 10 data objects in the consumer data set to be analyzed and uses them as the initial cluster centers;

[0039] (22) Calculate the distances between the 35,311 data sets and the 10 cluster centers respectively;

[0040] (23) For 35,311 data sets, each data set is assigned to the cluster with the highest similarity based on its similarity to the center of its cluster. The closer the distance, the higher the similarity.

[0041] (24) Calculate the average value of consumer data in each sub-market as the new cluster center;

[0042] (25) Preliminary market segmentation and the process of clustering consumer data are completed.

[0043] (3) Based on the clustering results obtained, three-branch clustering method is used to construct different consumer market spherical models to obtain market segmentation results. The following steps are included:

[0044] (31) The center of the sphere is: The radius of the sphere is: r i = max{dist(x,c i )|x∈C i}, construct the initial consumer sub-market sphere model; the center of the sphere is the cluster C i The average value of the consumer data set in the cluster, i = 1, ..., 10; the radius of the sphere is the distance from the farthest data set in the cluster to the center of the sphere; |C i | is the current ball C i The number of consumer samples in, x is the current particle C iConsumer data samples in, dist(.) represents the vector x and c i The Euclidean distance between them, x represents the current consumer sub-market particle C i Consumer data sample in, c i Represented as a sphere C i The center of the sphere;

[0045] (32) Set the number of algorithm iterations n, the weight of the data set in the core domain of the sphere w, and the weight of the data set in the boundary domain 1-w, where 0 <w<1;

[0046] (33) Calculate the distance between the center of the sphere in the sphere model of the i-th initial consumer sub-market and the spheres of other sub-markets;

[0047] (34) Determine the neighboring spheres; the details are as follows: Let the radius of the submarket sphere model C be r and the center be c; if C i is the neighboring sphere of sphere C, then Among them, r = max{dist(x,c)|x∈C}, x is the consumer sample in the current cluster. Then Make Assume that x is on line segment cc i On, we have ||xc||+||xc i ||=||cc i ||; That is, ||xc||>||xc i ||; That is, if C i is the neighboring sphere of sphere C, then the consumer data in sphere C is adjusted to sphere C i The possibility of

[0048] If C j is not a neighboring sphere of sphere C, then In the submarket sphere model C, ||cx||≤r, ||cx||≤||xc j ||;

[0049] (35) Determine the stable domain and the active domain; the details are as follows: Given any sphere C i , the set of its neighboring sphere centers is {N Ci};like Then for any sphere C in the sphere center set j , center c j ∈{N Ci}, with c i is the center of the ball, is the stable domain covered by the radius, and the particle C iThe area outside the stable domain is defined as the active domain; if Then the whole sphere C i All are stable domains.

[0050] (36) Update the sphere C for the stable domain i The center of the sphere c i ' and radius r i ', the formula is as follows:

[0051] r i '=max{dist(x,c i ')|x∈Co(C i )∪Fr(C i )'}, where |Co(C i )| represents the particle C i The number of consumer datasets in the core domain, |Fr(C i )'| represents the number of data sets in the new boundary domain after the consumer data sets in the original boundary domain are redistributed to the nearest neighbor, w represents the weight of the data set in the domain, and x represents the consumer data sets contained in the domain; the sample data in the remaining areas constitute the boundary domain and are distributed to the remaining sub-market spheres according to the principle of proximity.

[0052] Among them, the principle of proximity is as follows: i In the above example, after removing the data in the stable domain, the remaining data constitute the boundary domain of the sphere; the distances from the data in the boundary domain of the sphere to the centers of the 10 spheres are calculated in sequence, and the data are redistributed to the sphere model closest to it.

[0053] (37) Repeat steps (33) to (36) k times and one iteration is completed. If the centers of all spheres c i If there is no change or the number of iterations reaches the preset threshold n, then extract Get the ball C i The core domain of Get the ball C i The final market segmentation result is: C = {(Co(C1), Fr(C1), (Co(C2), Fr(C2)…, (Co(C k ),Fr(C k )}.

[0054] The embodiment of the present invention further provides a three-branch clustering system based on a particle sphere, comprising:

[0055] Acquisition module: used to use Matlab to obtain the information in the required market segmentation data set Pricerunner; the file format is mat;

[0056] Clustering module: used to analyze and cluster consumer data using the k-means algorithm;

[0057] Three-branch clustering module: used to construct different consumer market granular models based on the obtained clustering results using the three-branch clustering method to obtain market segmentation results.

[0058] An embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the computer program implements any one of the particle-sphere-based three-branch clustering methods described above.

[0059] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any one of the three-branch clustering methods based on particle spheres.

Claims

1. A three-branch clustering method based on granular spheres, characterized in that: The following steps are involved: (1) Use Matlab to obtain the information in the required market segmentation data set Pricerunner; the file format is mat; (2) Use the k-means algorithm to analyze and cluster consumer data information; (3) Based on the clustering results obtained, the three-branch clustering method is used to construct different consumer market granular models to obtain market segmentation results.

2. A three-branch clustering method based on spheres according to claim 1, characterized in that: In step (1), the information in the Pricerunner dataset includes: number of samples 35311, number of features 7, and number of categories 10.

3. A three-branch clustering method based on particle spheres according to claim 1, characterized in that: Step (2) is as follows: Using the k-means algorithm, the consumer information dataset U = {x1, x2, ..., x 35311 } Perform a preliminary clustering with an iteration number of 1 to obtain the preliminary market segmentation result C = {C1, C2, ..., C 10 }; In the clustering results, any two parts of the cluster C i With C j does not contain the same data set, and any cluster is not an empty set.

4. The three-branch clustering method based on particle sphere according to claim 1, characterized in that: Step (3) comprises the following steps: (31) The center of the sphere is: The radius of the sphere is: r i = max{dist(x,c i )|x∈C i }, construct the initial consumer sub-market sphere model; the center of the sphere is the cluster C i The average value of the consumer data set in the cluster, i = 1, ..., 10; the radius of the sphere is the distance from the farthest data set in the cluster to the center of the sphere; |C i | is the current ball C i The number of consumer samples in, x is the current particle C i Consumer data samples in, dist(.) represents the vector x and c i The Euclidean distance between them, x represents the current consumer sub-market particle C i Consumer data sample in, c i Represented as a sphere C i The center of the sphere; (32) Set the number of algorithm iterations n, the weight of the data set in the core domain of the sphere w, and the weight of the data set in the boundary domain 1-w, where 0 <w<1; (33) Calculate the distance between the center of the sphere in the sphere model of the i-th initial consumer sub-market and the spheres of other sub-markets; (34) Determine the neighboring spheres; (35) Determine the stable domain and active domain; (36) Update the sphere C for the stable domain i The center of the sphere c i ' and radius ri', the formula is as follows: r i '=max{dist(x,c i ')|x∈Co(C i )∪Fr(C i )'}, where |Co(C i )| represents the particle C i The number of consumer datasets in the core domain, |Fr(C i )'| represents the number of data sets in the new boundary domain after the consumer data sets in the original boundary domain are redistributed to the nearest neighbor, w represents the weight of the data set in the domain, and x represents the consumer data set contained in the domain; The sample data in the remaining areas constitute the boundary domain and are allocated to the remaining sub-market spheres based on the principle of proximity. (37) Repeat steps (33) to (36) k times and one iteration is completed. If the centers of all spheres c i If there is no change or the number of iterations reaches the preset threshold n, then extract Get the ball C i Extract the core domain of Get the ball C i The final market segmentation result is: C = {(Co(C1), Fr(C1), (Co(C2), Fr(C2)…, (Co(C k ),Fr(C k )}.

5. The three-branch clustering method based on particle sphere according to claim 1, characterized in that: Step (34) is as follows: Assume that the radius of the submarket sphere model C is r and the center of the sphere is c; if C i is the neighboring sphere of sphere C, then Among them, r = max{dist(x,c)|x∈C}, x is the consumer sample in the current cluster. Make Assume that x is on line segment cci, then ||xc||+||xc i ||=||cc i ||; That is, ||xc||>||xc i ||; That is, if C i is the neighboring sphere of sphere C, then the consumer data in sphere C is adjusted to sphere C i The possibility of If C j is not a neighboring sphere of sphere C, then In the submarket sphere model C, ||cx||≤r, ||cx||≤||xc j ||.

6. A three-branch clustering method based on particle spheres according to claim 1, characterized in that: Step (35) is as follows: Given any ball C i , the set of its neighboring sphere centers is {N Ci };like Then for any sphere C in the sphere center set j , center c j ∈{N Ci }, with c i is the center of the ball, is the stable domain covered by the radius, and the particle C i The area outside the stable domain is defined as the active domain; if Then the whole sphere C i All are stable domains.

7. The three-branch clustering method based on particle sphere according to claim 1, characterized in that: In step (36), the proximity principle is as follows: i In the above example, after removing the data in the stable domain, the remaining data constitute the boundary domain of the sphere; the distances from the data in the boundary domain of the sphere to the centers of the 10 spheres are calculated in sequence, and the data are redistributed to the sphere model closest to it.

8. A three-branch clustering system based on spheres, characterized in that: include: Acquisition module: used to use Matlab to obtain the information in the required market segmentation data set Pricerunner; the file format is mat; Clustering module: used to analyze and cluster consumer data using the k-means algorithm; Three-branch clustering module: used to construct different consumer market granular models based on the obtained clustering results using the three-branch clustering method to obtain market segmentation results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, the three-branch clustering method based on a particle sphere is implemented according to any one of claims 1 to 7.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a three-branch clustering method based on a particle sphere is implemented according to any one of claims 1 to 7.