Penalized Multinomial Mixture Model with Feature Selection for Text Clustering and its Application on Internet Public Opinions
Wei Xue · Tongji yu xinxi luntan · 2012
It is the key point that feature selection on high-dimensional sparse text data during the clustering of internet public opinions.By giving heavier penalty to the features having no significance to the result of clustering,penalized multinomial mixture model based on norm penalty idea can select the typical phrases on behalf of various types of public opinions effectively.This model has relatively better performance than the empirical research.