The amount of available data is increasing rapidly each year. The textual data holds a lot of information useful for several applications. In this talk we will present text mining techniques in order to select features in documents. For instance, the use of contextual and statistical information (e.g. web-mining) enables to identify Named Entities or to construct opinion lexicons. But words are not always relevant features. So a part of our work addresses terminology extraction, i.e. identification of relevant group of words. This talk will focus on the study of a specific term, i.e. the Title. Actually the automatic generation of titles is a complex task because titles have to be coherent, grammatically correct, informative, and catchy.