SUPPORT DATA MACHINES AND SEMANTIC KERNELS FOR TEXT CATEGORIZATION
Dr Florence d'Alché-Buc, Thème Apprentissage et Acquisition des connaissances Laboratoire d'Informatique de Paris 6 - LIP6

Abstract: The learning principle of Support Vector Machines can be applied to any data (not only vectors) that can be described as proximity data by means of a kernel. In many pattern recognition applications, this can lead to consider using more expressive representations of the data, eventually structured ones, combined with well-founded statistical learning tools. Text categorization and, more generally, information retrieval tasks offer a very promising application domain for testing these ideas. We explore and compare different ways of representing documents relying on kernels that encode semantic relationships between words.

Results in some text categorization tasks are presented and commented.

This seminar was held at the Department of Computer Science, Royal Holloway, University of London on 9 October 2000.

back