Feature Weighting Information-Theoretic Co-Clustering for Document Clustering
Yunming Ye, Xutao Li, Biao Wu, Yan Li · 2009
This paper presents a feature weighting schema to improve the performance of Information-Theoretic Co-clustering (ITCC). The new algorithm, named as Feature Weighting Information-Theoretic Co-clustering (FWITCC), weights each feature with the mutual information shared by the features and the documents. The weighting schema makes informative features more important and noisy features less important, so that it can improve the qualities of resulted clusters. Experimental results on both synthetic data sets and 20Newsgroup data sets have demonstrated that our new approach has better clustering performance than (ITCC).