基于TFIDF+LSA算法的新闻文本聚类与可视化

News Text Clustering and Visualization Based on TFIDF+LSA Algorithm

近几年来,文本聚类技术作为机器学习领域一种无监督学习的方法,也越来越成为数据挖掘领域备受关注的技术之一。将小规模的文本数据聚为几类,在一定程度上说是一件比较容易实现的工作。可是,当面对大量高维的中文文本数据时,由于在这种情况下对文本聚类,面对的将是高维和稀疏的数据,在保证聚类质量的情况下,提高聚类的速度和可视化效果也成为聚类研究的课题之一。该文提出一种结合词频反文档频率算法(term frequency,inverse document frequency,TFIDF)和潜在语义分析算法(latent semantic analysis,LSA)相结合的方法,来提高kmeans中文文本聚类的速度和可视化效果。将从网页上采集到的11456条新闻作为实验对象,通过基于TFIDF聚类和基于TFIDF+LSA聚类进行实验对比,根据聚类指标轮廓系数(Silhouette coefficient,SC)、卡林斯基-原巴斯指数(Calinski-Harabasz index,CHI)和戴维斯-堡丁指数(Davies-Bouldin index,DBI)的值表明,该方法不仅能保证文本聚类的质量,还能大大提高文本聚类的速度和可视化效果。

In recent years,as an unsupervised learning method in the field of machine learning,text clustering technology has increasingly become one of the most concerned technologies in the field of data mining.To a certain extent,it is a relatively easy work to aggregate small-scale text data into several categories.However,when faced with a large number of high-dimensional Chinese text data,text clustering in this case will be faced with high and sparse data,while ensuring the quality of clustering,improving the clustering speed and visualization effect has become one of the topics of clustering research.We propose a method combining term frequency inverse document frequency(TFIDF)algorithm and latent semantic analysis(LSA)to improve the speed and visualization of kmeans Chinese text clustering.In this paper,11456 pieces of news collected from web pages are taken as experimental objects,and the experimental comparison is made based on TFIDF clustering and TFIDF+LSA clustering.According to the clustering index like Silhouette coefficient(SC),Calinski-Harabasz index(CHI)and Davies-Bouldin index(DBI),the proposed method can not only guarantee the quality of text clustering,but also greatly improve the speed and visualization of text clustering.