Massive use of digital and web documents makes text categorization
an important issue as one would rather retrieve documents that match
perfectly one’s search criteria. A new categorization technique is proposed.
Proposed method is intended to offer a classical text categorization
engine developed through the means of machine learning techniques. Experiments
were conducted on the Reuters-21578 text categorization test
collection. The proposed method correctly predicts 73% of text assignations.
A hierarchical representation of the resulting assignation is obtained
by using equivalence classes in an information system. This can be further
used in building more precise search engines.