An Effective Sub-space Learning for Document Clustering

Clustering in textual document attracts more and more attention with the huge Internet news data appears every day. But the performance can be in∞uenced by the high dimension vectors based on wordbag. Many of which are redundant information. Hence, it is very necessary to derive a low dimensional subspace that contain less redundant information in order to make the documents can be clustered more reasonably. In traditional studies, learning a subspace and clustering vectors are divided into two independent steps, e.g., feature selection step and clustering step. In this way, it could not estimate whether the subspace is appropriate for the clustering. To solve this issue, we select the feature flrstly, and combine the subspace learning step and the clustering step into an iterative procedure. First, we take into account similarity between the intra clusters and separability in the inter cluster documents, and then utilize the a‐nity propagation to adopt to partition the number of clusters. The experimental results show that our proposed method outperforms the conventional methods of document clustering using our data set.