December 06, 1999
10:30am

Fernanda Caropreso
School of Information Technology and Engineering
University of Ottawa

Experiments in Text Categorization with Statistical Phrases

Previous research in Text Categorization shows that the use of n-grams may improve the obtained result. However, usually these n-grams have been obtained by a simple frequency-based feature selection method. In our experiments we use more sophisticated pruning techniques that take the class information into account.

The presentation describes:

• the way we obtain the n-grams and the experimental setup,
• the Ripper rule learning algorithm used,
• the new results compared with the previous ones,
• the planned future work.