A text categorization framework based on concept structure

From the view of concept structure in psychology, this paper analyses the cause of classes overlapping of training sample set in text categorization, and finds it is reflection about fuzzy boundary of nature concept in space distribution between training samples. The paper proposes a categorization framework based on concept structure. In the framework, concept structure and relation between concepts are shown more obviously through anew partitioning training sample set, and then based above processing results we construct hierarchy classifier to improve classification performance. The experiments show that the framework is effective for improving classification performance and its expansibility is good.