The Research Of Social Network Customer Segmentation Based On Clustering Analysis And Decision Tree Algorithm

In recent years,driven by the mobile Internet,the user groups of social networks have been expanding.At the same time,the differences between different user groups are increasingly obvious.As the most representative large scale social platform in China,Sina weibo has a large number of users expressing their opinions on various things through microblog every day.For this reason,this thesis selects the blog posts of Sina weibo users as the research object,hoping to find out the valuable information hidden behind the views expressed by users and classify users' emotional tendency through effective methods.This is not only conducive to enterprises' "right medicine" and precision marketing,but also conducive to the government's understanding of the public views on a macro level,providing a basis for the formulation and implementation of various policies.In October 2015,China implemented a two-child policy.Most scholars have studied the views and emotional tendencies of Chinese people on "second child" from the perspectives of sociology and demography.This thesis will use text mining algorithm to deeply analyze this problem on the basis of relevant theories.Firstly,the thesis collects 5303 blog posts with "two-child" as the key word in Sina weibo by network capture method,and realized text denoising,Chinese word segmentation,word frequency statistics,text representation and other preprocessing work on the data.Then,k-means and k-mediods are used for cluster analysis of user opinions,which are divided into positive,neutral and negative categories,accounting for 13.1%,76.6% and 10.3% respectively.Then,the decision tree and k-nearest classifier are established respectively with the clustering results as the category characteristics.Finally,the comparison results show that the micro-precision and micro-recall rates of the decision tree model are both 73.4%,and the micro-precision and micro-recall rates of the k-nearest model are both 59.3%.The decision tree has stronger fitting ability and can effectively realize the classification of short texts on weibo.