Social networks have a community providing feedback on comments that allows to identify opinion leaders and users whose positions are unwelcome. Other platforms are not backed by such tools. Having a picture of the community's reactions to a published content is a non trivial problem. In this work we propose a novel approach using Abstract Argumentation Frameworks and machine learning to describe interactions between users. Our experiments provide evidence that modelling the flow of a conversation with the primitives of AAF can support the identification of users who produce consistently appreciated content without modelling such content.