•1,Inthe first phase, we determine the best topic model and present a formal conceptabout significance degree of topics and some topic selection criteria, throughwhich we can find the best number of the most suitable topics from the originaltopic model discovered by LDA.
•在这一步中,我们提出了主题重要度的概念和主题选择标准,并由此得到表示文档集的最佳主题个数。
•2,Thenwe choose the initial clustering centers by using k-means++ algorithm.
•第二步,利用k-means++算法确定k-means算法初始的中心点。
•3,Inthe third phase, we take the obtained initial clustering centers and usek-means algorithm for document clustering.
•第三步,我们利用第二步得到的中心点作为k-means算法的初始点,进行文本聚类。
K-means++clustering
•K-means++算法提供了一种为k-means算法选取初始中心点的方法。
Determining the Best Topic Model
•1,SignificanceDegree of Topic ti withrespect to Document dj,Sig(dj,ti);
Determiningthe Best Topic Model
•SignificanceDegree of Topic ti Sig(ti)
TopicSelection Criterion
•我们将从LDA模型中得到的N的主题,按照他们的Sig(t)的大小,从大到小的顺序进行排列。然后找到一个M,满足