Profiling Twitter users with diffusions

This vignette gives a tutorial on profiling Twitter users with their initiated diffusions via modeling methods provided by evently. We use the dataset auspol shipped with evently in this task. It is a list of data frames (3333 in total) where each data frame represents a diffusion cascade. For each cascade, three fields are presented: time is the retweeting time relative to the original tweet (i.e., the first event); user is the corresponding Twitter user of the event; magnitude is the number of followers the Twitter user has.

head(auspol, n = 3)
#> [[1]]
#>   magnitude time user
#> 1        58    0 2984
#> 
#> [[2]]
#>   magnitude time user
#> 1    298956    0  630
#> 2     10414 1124 1901
#> 3      3099 1329 3566
#> 4      1947 5822  490
#> 
#> [[3]]
#>   magnitude time user
#> 1      1627    0 2692

names(auspol) <- sapply(auspol, function(data) data$user[[1]])
auspol_group_fits_by_user <- group_fit_series(data = auspol, model_type = 'mPL', observation_times = Inf, cores = 10)

head(auspol_group_fits_by_user, n = 3)

# keep only users with more than 5 cascades
auspol_group_fits_by_user <- Filter(function(x) length(x) >= 5, auspol_group_fits_by_user)
dist_matrix <- fits_dist_matrix(auspol_group_fits_by_user)

positions <- tsne(dist_matrix, k = 2)

ggplot(as.data.frame(positions), aes(V1, V2)) +
  geom_point() +
  theme_void()

Quyu Kong

2021-01-30