Fig 1.
Monthly changes of blogs over 10 years.
(A) Each emotional dynamics Zk(t) Tension, Depression, Anger, Vigor, Fatigue, and Confusion are shown from top to bottom. (B) Standardized numbers of summed emotions (top) and the whole number of blogs X(t) (bottom).
Fig 2.
Weekly and yearly periodicities p(l).
(A) Weekly periodicities p(l) for each of the emotional dynamics with error bars representing standard deviations, s(l), over 520 weeks. Most emotions show differences between weekdays and weekends. (B)(C) Yearly periodicities p(l) for each of the emotional dynamics in monthly scale (bold) and in daily scale (dotted) with shaded area representing standard deviations s(l) for nine years. Fatigue increases during summer times (July and August) while Depression and Confusion slightly increases during winter times (December and January).
Table 1.
Major dates in which emotion increased significantly every year.
Rates are calculated from the temporal average, see also Fig 2.
Table 2.
Major spikes in descending order of increased rate which are estimated from averaging earlier seven days.
Tension shows many spikes due to earthquakes and typhoons.
Fig 3.
Examples of sharp spikes of collective emotion after removing weekly and yearly periodicities.
(A) Tension, Depression, and Confusion continue to increase for more than a week at the 3.11 earthquake period in 2011. (B) Tension at the Kumamoto (the southwest of Japan) earthquake in 2016, and (C) Confusion at the heavy snowfall in the Tokyo metropolitan area in 2016. Except for the 3.11 earthquake, most sharp spikes returned to their original baseline within a week.
Fig 4.
Autocorrelation functions ρk(τ) and power spectral density Sk(f) of collective emotion.
ρk(τ) and Sk(f) are calculated for (A)(B) Raw emotional dynamics which shows periodic cycles, (C)(D) Emotional dynamics without periodic cycles which shows long-term memory, (E)(F) Weekly randomized dynamics which diminished correlations larger than weekly scale.
Fig 5.
Results of principal component analysis for each emotion time series organized every six months.
Results of (A) Raw emotional dynamics, that moved gradually for every six month, and (B) Weekly randomized dynamics that jumped from point to point are shown.