Figure 2. Each cluster represents a group of language editions exhibiting a similar temporal trend, according to the execution of the k-means clustering algorithm. Time is expressed on the x-axis in the number of months since the creation of a language edition. Gray lines represent the time series of the individual language editions, and red lines represent the average over each cluster. The language editions belonging to each cluster are reported below, with background indicating the size range in the number of active editors.
Figure 3 shows more in detail the evolution of these timelines (number of monthly active editors over time) for the sample of 8 selected languages that we will consider from now on for presenting the results. Apart from more or less marked seasonality patterns (with peaks of lower activity in summer), the graphs show for English and German a clear pattern of rapid growth until 2007, and then slow decline until more or less stabilizing in recent years. These are indeed the major languages in cluster 0; Polish, also in cluster 0, presents a similar pattern, although decline starts a bit later and presents a smoother trend. We see the same rapid growth in the first years for the Italian Wikipedia, followed by a smooth decline and then a smooth growth. For the Arabic Wikipedia, we can observe more in detail the sustained growth pattern of cluster 5. Catalan exhibits a rapid growth initially, followed by a stable pattern, apart from seasonal oscillations. Finally, Afrikaans presents growth until 2009, then a mostly stable pattern until a smooth growth in recent years, while Swahili reaches a similar number of active editors, but growing primarily after 2017.
