Fig 1.
Circadian patterns of the Sina Weibo Hot Search List (HSL).
(A) Increment of number of new hashtags per Δt = 5 minutes on the HSL during the observation period from 22 May 2020 to 29 September 2020. (B) Time series of the median of search volume index of all hashtags on the HSL at a timestamp, advertisement rank positions excluded. represents the median value hotness H of hashtags on Sina Weibo HSL at a timestamp. In both (A) and (B) the one-week gap due to the suspension of HSL by the cyberspace authority of China is visible. An enlarged part of (A) is in S1 Appendix.
Fig 2.
Clustering patterns of hashtag rank trajectories on the Sina Weibo HSL.
(A) Distribution of hashtag duration on the HSL, divided into two sections based on local minima at 1 hour. Results of k-means clustering with 3 clusters in each section for time series data are shown, metric is Dynamic Time Warping (DTW) distance, y-axis is normalized to the mean and the standard deviation and the x-axis by di. (B), (C), (D) correspond to duration interval from 0 to 1 hour (Section 1). (E), (F), (G) correspond to duration interval larger than 1 hour (Section 2). Red curves depict clustering centers (centroid) [39], computed as the barycenters [40] with respect to DTW. (We performed the clustering also with 4 clusters for both categories, see S1 Appendix).
Fig 3.
Relationship between hashtags’ duration on the HSL and the time ti.
(A) Scatter plot of hashtags’ duration on the HSL and the time of the day they first appear on the HSL. Each point is a hashtag, colored by the category it is clustered in Fig 2. (B) Distribution of hashtags’ duration on the HSL according to different time intervals during the day of first appearance on HSL.
Fig 4.
Ranking dynamics characterization of hashtags on the Sina Weibo HSL from 17 July 2020 to 17 Sep 2020.
(A) Distribution of ri(ti) and ri(Ti). (B) Scatter plot of and di. (C) Scatter plot of ri(ti) and di, hashtags with high enter-rank and short duration are circled red. (D) Scatter plot of ri(Ti) and di, rank 33 marked by red arrow.
Fig 5.
Parabola shaped normalized rank diversity in a model closed system of size 500.
The top L = 48 can be considered as an open system.
Fig 6.
Prehistory length tHSL, enter-ranks ri(ti), the highest rank , and duration di of hashtags on the Sina Weibo HSL.
(A) The relationship between the hashtags’ prehistory time length and the ranks they first enter on the HSL. (B) The relationship between the hashtags’ prehistory time length and the highest rank during stay on the HSL. (C) The relationship between the hashtags’ prehistory time length and the duration they stay on the HSL. (D) Parameterized probability density function of the hashtag duration on the HSL by prehistory time length, using kernel density estimation (KDE) [41], with the parameter bw = “scott” [42].
Fig 7.
Rank dynamics comparison between empirical data and a ranking model with anchoring.
(A) Empirical rank diversity separated for day (upper line) and night (lower line). The sudden drops are at ranks 8, 16, 28, and 33. (B) Simulated rank diversity with the anchor effect.
Fig 8.
Categorized proportion of hashtags that have stayed at certain ranks on HSL for longer than 2 hours.
(A)(B)(C)(D) show the content distribution of hashtags at ranks 8, 16, 28, and 33 respectively, corresponding to the sudden drops in Fig 7A. (E) Averaged proportion of hashtags by content category at ranks 5, 12, 21, 25, 30, 37.