Adaptive Navigation systems (ANS) are intended to support user navigation in a hypermedia system. In the context of the Web, ANS suffer from the difficulty of finding relevant criteria to represent the users' current information needs and their frequent shifts of focus. This paper addresses the issue of user's information needs modeling. The proposed approach is to find relevant clues about the user's current interests from the content of the accessed pages. First, the concept of clue about user's information needs is formalized. Secondly, a clue extraction algorithm in a stream of accessed pages is presented. Then, VISS, a visualization tool intended to display user's different search activities with respect to clues found in a stream of accessed pages is put forward. Finally, this paper discusses the result of a Web navigation survey intended to characterize user's shifts of focus.