In this chapter, we discussed all the basic NLP techniques, starting from the definition of a corpus up to the final transformation into feature vectors. We analyzed different tokenizing methods to address particular problems or situations of splitting a document into words. Then we introduced some filtering techniques that are necessary to remove all useless elements (also called stopwords) and to convert the inflected forms into standard tokens.

These steps are important in order to increase the information content by removing frequently used terms. When the documents have been successfully cleaned, it is possible to vectorize them using a simple approach such as the one implemented by the count-vectorizer, or a more complex one that takes into account the global distribution of terms, such as tf-idf. The latter was introduced to complete the work done by the stemming phase; in fact, it's purpose is to define vectors where each component will be close to 1 when the amount of information...