In this paper, we propose a general visual and interactive tool named Donatien that supports the comparison of collections of doc- uments like sets of documents evolving over time or query results coming from different sources. Our approach builds on previous work from differ- ent areas of research and conciliates visualization, interaction and opti- mization approaches. Sets of documents are represented by graphs where documents are nodes, and edges between nodes are relationships between documents. Our contribution is to combine (1) a layer mechanism that makes possible the superposition of different layers representing differ- ent sets of documents, (2) original deterministic layout strategies that facilitate comparisons across layers, and (3) an interaction model based on drag-and-drop that lets users manually adjust matchings between documents. Our approach is used in a real case study based on data ex- tracted from the InfoVis contest benchmark. The data used contains ap- proximately 600 research papers with approximately 8000 citation links between documents covering almost 10 years of the InfoVis conference. Different subsets, corresponding to different years, are extracted from the data and compared with our approach in order to visually analyze the evolution of papers and topics over years.