Class-driven correlation learning for chinese document categorization using discriminative features
Xian Wei Wu, Lingli Zhou, Xiang Li, Jianhuang Lai · 2011
This paper proposes a class-driven correlation learning method for Chinese document categorization to assign one suitable category in the first level to a document. Discriminative features are selected from candidate terms with high occurrence probability in each category. Class-driven correlation learning is then performed to produce a set of projections and further construct a code matrix to record the correlations between different classes of documents. A new document is classified by implementing the decision rule through the results from class-driven correlation learning. The competitive results from the experiments performed on TanCorp corpus indicate the superiority of the proposed method.