Micro-blogging systems have become a prime source of information. However, due to their unprecedented success, they have to face an exponentially increasing amount of user-generated content. As a consequence finding users who publish quality content that matches precise interest is a real challenge for the average user. This paper presents a new recommendation score which takes advantage both of the social graph topology and of the existing contextual information to recommend users to follow according to user interest. Then we introduce a landmark-based algorithm which allows to scale. The experimental results and the user studies that we conducted confirm the relevance of this recommendation score against concurrent approaches as well as the scalability of the landmark-based algorithm.