Tweets exchanged over the Internet represent an important source of information, even if their characteristics make them difficult to analyze (e.g. a maximum of 140 characters, etc.). In addition associated to every message, lots of information such as location or date, are available. Taking into account these meta-information can be very useful for the decision maker. Obviously, due to the characteristics of tweets, relevant information are not expressed. For instance when considering tweets about natural disasters, an automatic extraction of location can significantly improve the analysis. In this paper, we propose a new approach to automatically extract and exploit the information about location. We also propose the end-user with a geographical hierarchy in order to improve the analysis of a set of tweets. Experiments carried out on real medical data underline the relevance of our proposal.