Text Categorization with Modified LSI

Automatic Text Categorization (TC) is a complex and useful task for many naturallanguage applications, and is usually performed through the use of a set of manuallyclassified documents, a training collection. Term-based representation of documents hasfound widespread use in TC. However, one of the main shortcomings of such methods isthat they largely disregard lexical semantics and, as a consequence, are not sufficientelyrobust with respect to variations in word usage. We shall design, implement, and evaluatea new, text classification algorithm. Our main idea is to find a series of projections ofthe training data by using a new modifided LSI algorithm, project all training instancesto the low-dimensional subspace found in the previous step, induce a binary search onthe projected low-dimensional data. Our conclusion is that, with all its simplicity andefficiency, our approach is comparable (and sometimes superior) to SVM in terms ofaccuracy