Fig. 5

Mean entropy 〈H_i〉 and 95% CI of topics in (a) posted and (b) received tweets over sliding windows with a size of k=10. CI become wider due to the decreased sample sizes of users having a large number of tweets. Legends report estimated coefficients B and p-values in a linear regression model: 〈H_i〉=Bi+ε, where estimations are based on windows over i∈[1,150] due to relatively small mean errors in this range