Using bigrams detection for text categorization in scientific domain

This paper presents some experiments using the technique of multi-words detection for text categorization in scientific domain. We have used part of the collection of scientific papers of High Energy Physics (HEP) provided by the European Laboratory for Particle Physics (CERN). The supervised machine learning algorithms employed have been Rocchio and PLAUM. The technique of multi-words detection used has been limited to fixed sequences of maximum two terms, known as bigrams. The aim of this study is to determine whether the use of frequent bigrams as unique features may be an improvement for text categorization task in this specific domain. Our conclusion is that multi-words detection should not be used for this task in the HEP domain.