Social networks have become a primary communication tool and are used by hundreds of millions of users daily. They bring together a wide variety of people, individuals, companies, public figures, media, influencers, etc. Users have different behaviours on social networks, such as different publication frequencies, number of followers or different user interactions. In the Twitter social network, for instance, users do not reply, quote or use mentions in the same way. Our intuition is that these interactions may characterise different user types and we consequently present in this work a non-supervised classification method based on interaction scores. We propose and experimentally compare different score estimations, leading our experiments to confirm the relevance of our approach.