Figure 4
From: Estimating tie strength in social networks using temporal communication data
Exploratory analysis of features of human communication: (i) Feature correlation matrices measured by Pearson’s correlation coefficient. Features are sorted according to their modelling approach, each case divided by pink lines. Features display high within-group correlations, with lower between-group correlations. Weekly clusters (last group) show no relevant correlations among themselves or to other variables, with two main exceptions: negative correlations between clusters C1 (late night) and C5 (weekday worktimes), and clusters C5 and C9 (weekend night). (ii) Average topological overlap given the ranks of three variables correcting for three different levels of communication intensity (w), with the shaded area depicting 80% of the distribution. From top to bottom: number of bursty trains (\(N^{E}\)), Jensen-Shannon Divergence for difference in daily patterns (JSD), temporal stability (TS). Variable rankings are normalized to be on the \([0,1]\) interval