Representation of documents is the basis of clustering systems. In addition, non-contiguous phrases appear more and more frequent in the text in the Web 2.0 age, and these phrases can affect the result of text clustering. In order to improve the quality of text clustering, this paper proposed a feature cluster-based vector space model (FC-VSM) which used the text feature clusters co-occurrence matrix to represent document and proposed to identify non-contiguous phrases in the text preprocessing stage. Our method can reduce dimension of features compared with the traditional VSM-based model. It identified non-contiguous phrases, used distributed representation of features, and implements feature clusters. Despite their simplicity, our methods are surprisingly effective and can improve the accuracy of clustering significantly which is shown in experimental results.
展开▼
