Using Multiple Sets of Attributes for Text Categorization
Yaxin Bi, Qiang Zhang, Shengli Wu, Jiwen Guan · 2006
This paper investigates how multiple sets of attributes can be generated using a rough sets-based inductive learning method and how they can be combined for improving classification decisions, particularly in the context of text categorization, by using Dempster's rule of combination. We first propose a boosting-like technique for generating multiple sets of attributes based on rough set theory, and a method for transforming multiple sets of attributes to multiple sets of rules, and then model classification decisions inferred by the rules as pieces of evidence. The various experiments have been carried out on 10 out of the 20-newsgroups - a benchmark data collectionndividually and in combination. Our experimental results support the claim that "decisions made by multiple experts would be more effective than any one if their individual judgments are appropriately combined"