Font Size: a A A

Research On Customer Classification Model And Improved Particle Clustering Algorithm

Posted on:2019-11-06Degree:MasterType:Thesis
Country:ChinaCandidate:S Y HuangFull Text:PDF
GTID:2428330566983526Subject:Software engineering
Abstract/Summary:PDF Full Text Request
With the development of Internet technology and express logistics,online shopping has gradually become a new shopping habit for people in China.The scale of China's online shopping market has become unprecedentedly large.As enterprises attract more and more customers to the online shopping platform,how to use valuable and limited service resources to provide high-quality services to customers is still in a dilemma.Only by classifying the huge customers by groups effectively,the limited service resources of enterprises could be focused on high-quality customers and to maximize the business interests.In addition,the large number of customer data accumulated by enterprises themselves has created possible for the customer classification and then to serve highquality customers.The customer value analysis models represented by RFM and some clustering algorithms are widely used customer classification technologies.However,the existing customer classification models and clustering technologies still have problems such as too single model indexes,too detailed classification,sensitive to outliers and unstable clustering result.In view of the problems that the traditional customer classification model has too single indexes and the customer classification is too detailed,this paper proposes a CCL customer classification model,including C(Customers),C(Contribution),L(Loyalty)these three model indexes.The three indexes in the model are no longer just only a single dimension,but rather cover the customers' static and dynamic attributes.Firstly,according to the significance of each index,the attributes of multiple dimensions in the customer data are selected,so as to avoid the single problem of indexes.Secondly,the indexes are obtained by means of different dividing and quantification standards,characteristic value structure and weighted calculation of comprehensive score.Finally,this paper combines CCL model and clustering algorithm to classify customers more effectively.K-means clustering algorithm is limited by K value selection and initial cluster centers,and outliers in the cluster samples will result in large deviation of clustering results,resulting in extremely unstable clustering results.Aiming at the existing problems,this paper proposes a new particle swarm optimization algorithm based on K-Medoids.In this paper,using the K-Medoidsclustering algorithm solves the K-Means suffers from the problem of outliers,and the improved particle swarm algorithm introduced in Opposition-based Learning strategy,through repeated iteration and tweaking in the larger search space,get the global optimal cluster centers.Finally,the paper combines the CCL customer classification model and the improved particle swarm algorithm based on K-Medoids to form an improved particle swarm clustering algorithm based on CCL model.Not only the indexes of CCL model are comprehensive and representative,but also K-OD-PSO algorithm solves the problems that the traditional algorithm is sensitive to outliers,easy to fall into the local range of the optimal solution,and the rapid but excessive convergence.Therefore,the algorithm has a good effect of customer classification.In this paper,in the design of experiment environment,an improved particle swarm clustering algorithm based on CCL model is proposed,in order to evaluate the clustering algorithm of clustering,classification,performance and stability classification efficiency.After repeated experiments,it is proved that the improved particle swarm clustering algorithm based on CCL model has excellent stability and classification performance,which satisfies the original intention of the algorithm.
Keywords/Search Tags:Customer classification model, Clustering algorithm, Particle Swarm Optimization, Opposition-based Learning
PDF Full Text Request
Related items