In this paper, we introduce a transformation of time-varying graphs, called Delta-duplication, for the analysis of heterogeneous dynamic network. Instead of building a sequence of snapshots from non-overlapping time windows, we propose a relative approach : consider a vertex as different vertices over multiple sessions. A session being a time period where the vertex interaction are not separated by more than Delta timestamps. We describe here the general theory of ∆-duplication and provided some directions for applications to TVG analysis. In particular, we introduce a generalization of k-core to temporal graph using this model.