A New Text Categorization Method Based on SVD and Cascade Correlation Algorithm

A new text categorization method based on singular value decomposition (SVD) and cascade correlation (CC) algorithm is proposed. Most traditional classification systems represent the contents of documents with vector space model (VSM) which represents documents with a set of index terms. However, this model needs a high dimensional space to represent the documents and it does not take into account the semantic relationship between terms, which could lead to poor classification performance. In contrast, SVD can represent relations among very large number of words and very large number of natural text passages in which they occur. It can not only greatly reduce the dimensionality but also discover the important relationships between terms. Based on this idea, we use singular value decomposition (SVD) to represent our documents in this paper. Then we use neural network constructed by cascade correlation (CC) algorithm to classify these represented documents. The experiments show that our method helps to accelerate the training speed and improves the classification accuracy as well.