Mining opinion data that reside in online discussions is a way to track opinions of people on specific subjects. Many of the existing techniques model a discussion as a social network of users and they represent it with a user-based graph. In this paper we propose a new framework for discussion analysis. We combine Social Network Analysis and Opinion Mining in order to give structure to a discussion. Such techniques have not been combined until now.We propose the use of an opinion-based graph whose vertices contain message objects and its «reply-to» edges are labeled with opinion polarities. We compare the opinionbased with the user-based graphs and we analyze the different information that can be extracted from them. Our experiments validate the proposed framework and show that the representation of discussions by opinion-based graphs gives information that cannot be provided by a user-based graph.