In this paper, the problem of classifying a HTML documents into a hierarchy of categories is investigated in the context of cooperative information repository, named WebClassII. The hierarchy of categories is involved in all aspects of automated document classification, namely feature extraction, learning, and classification of a new document. Innovative aspects of this work are: a) an experimental study on actual Web documents which can be associated to any node in the hierarchy; b) the feature selection process; c) the automated selection of thresholds for the score returned by a classifier; d) the comparison of three different techniques (flat, hierarchical with proper training sets, hierarchical with hierarchical training sets); e) the definition of new measures for the evaluation of system performances. Results show that the use of hierarchical training sets improves the hierarchical techniques.
展开▼
