The layout is pre-computed so that streams topology is consistent with empirical semantic proximities. Clusters are then linked inter-temporally with streams, the closer two successive clusters, the darker they are. Clusters width also change according to the number of associated records.
This visualization then allows to follow both continuous dynamics (increase or decrease of streams width) and discrete dynamics (emergence, forking or merging events, loss or gain of terms, etc.).
Herebelow, an example is shown, it comes from the analysis of a scientific publications dataset about biofuel. A more informative interactive version is also available (click the image).