Figure 1.
Regular and irregular states of the tweet stream.
(A) Tweet count series for February and March 2011. Unthresholded recurrence plot [41] of the tweet time series (B) before the Tohoku-oki earthquake (March 4–10, 2011) and (C) after the earthquake (March 11–17, 2011). The gray scale corresponds to the distance between the tweet count series in a reconstructed phase space. Darker points indicate that two corresponding states are close to each other, whereas lighter points indicate that two corresponding states are distant from each other.
Figure 2.
Tweet probability distribution for March 11, 2011, the day of the Tohoku-oki earthquake, showing a burst-like increase and a sustained active state, whereas those for March 1 and 10 exhibit almost the same temporal profiles as those on normal days.
Figure 3.
Distributions of the JS values related to collective attention.
Intensity of collective attention (i.e., the value) in (A) 2010 and (B) 2011. (C) Distribution of
values on weekdays, weekends, and national holidays. Only
values
are shown.
Figure 4.
Examples of strong collective attention.
Tweet probability distribution, and the rankings of the popularity and the enhancement of key terms in Japanese tweets on (A) March 11, 2011 and (B) July 18, 2011. The collective attention in (A) was associated with the Tohoku-oki earthquake and tsunami, whereas that in (B) was associated with the FIFA Women's World Cup, 2011. The black bars in the figures indicate the time period when the posted tweets were analyzed to compute term frequency. The red text in the tables denotes terms related to the target events and the text in parentheses shows the English translation.
Figure 5.
Examples of moderate collective attention.
Tweet probability distribution, and the rankings of the popularity and the enhancement of key terms in Japanese tweets on (A) December 9, 2011 and (B) December 10, 2011. The collective attention in (A) was associated with the animated movie Castle in the Sky, whereas that in (B) was associated with a total lunar eclipse. The black bars in the figures indicate the time period when the posted tweets were analyzed to compute the term frequency. The red text in the tables indicates terms related to the target events and the text in parentheses shows the English translation.
Figure 6.
Visualization of RT networks during collective attention.
(A) Tohoku-oki earthquake ( of nodes = 27,340,
of links = 27,709), (B) FIFA Women's World Cup 2011 (
of nodes = 9,277,
of links = 8,450), (C) Castle in the Sky (
of nodes = 1,183,
of links = 793), and (D) total lunar eclipse (
of nodes = 893,
of links = 553). The nodes represent users, which are connected if there is a RT with key terms related to the target event. In each figure, the node sizes are proportional to the number of retweeted tweets. Only the nodes with
retweeted tweets are shown for clarity.
Figure 7.
Structural features of RT networks.
(A) Degree distribution of the RT networks related to the Tohoku-oki earthquake, FIFA Women's World Cup, 2011, Castle in the Sky, and total lunar eclipse. (B) Boxplots of the connected component sizes in the RT networks.
Figure 8.
Parameter dependence during the detection of collective attention candidates.
Smaller values of and
were correlated with the detection of more collective attention candidates. However, many false positive candidates were detected when these parameters were very small.