In Text Categorization (TC) based on Vector Space Model, feature weighting and feature selection are major problems and difficulties. This paper proposes two methods of weighting features by combining the relevant influential factors together. A TC system for Chinese texts is designed in terms of character bigrams as features. Experiments on a document collection of 71,674 texts show that the F1 metric of categorization performance of the system is 85.9%, which is about 5% higher than that of the well-known TF*IDF weighting scheme. Moreover, a multi-step feature selection process is exploited to reduce the dimension of the feature space effectively in the system.
展开▼
