In this informative age, we find many documents are available in digital forms which need classification of the text. For solving this major problem present researchers focused on machine learning techniques: a general inductive process automatically builds a classifier by learning, from a set of pre classified documents, the characteristics of the categories. The main benefit of the present approach is consisting in the manual definition of a classifier by domain experts where effectiveness, less use of expert work and straightforward portability to different domains are possible. The paper examines the main approaches to text categorization comparing the machine learning paradigm and present state of the art. Various issues pertaining to three different text similarity problems, namely, semantic, conceptual and contextual are also discussed.