The second half of my talk mostly focuses towards the detection and analysis of dynamical communities in social networks, specifically in citation network. Most of the recent methods aim at exhibiting community partitions from successive graph snapshots and thereafter connecting or smoothing these partitions using clever time-dependent features and sampling techniques. These approaches are nonetheless achieving "longitudinal" rather than 'dynamic' community detection. Assuming that communities are fundamentally defined by a certain amount of interaction recurrence among a possibly disparate set of nodes over time, we suggest that the loss of information induced by considering successive snapshots makes it difficult to appraise essentially dynamic phenomena. We propose a methodology which aims at tackling this issue in the context of citation datasets, and present several illustrations on both empirical and synthetic dynamic network datasets.