Sentiment of tax-related tweets sent to:realDonaldTrump

tax-tweets-to-trump.R

## install packages (from CRAN) if not already
pkgs <- c("dplyr", "rtweet", "ggplot2", "syuzhet", "ggbeeswarm", "remotes")
if (any(!pkgs %in% installed.packages())) {
  install.packages(pkgs[!pkgs %in% installed.packages()])
}

## install {dataviz} theme from github
remotes::install_github("mkearney/dataviz")

## define paste function I really like rn
`%P%` <- function(lhs, rhs) paste0(lhs, rhs)

## load rtweet & ggplot2
library(rtweet)
library(ggplot2)

## tax-related [english] tweets sent to:trump
yes_tax <- search_tweets(
  "to:realdonaldtrump (my taxes) OR (i tax return) OR (my tax return) " %P%
    "OR (our tax return) OR (i owe taxes) OR (my tax refund) lang:en",
  n = 5000,
  include_rts = FALSE
)

## filter first tweet from each user
yes_tax <- dplyr::filter(yes_tax, !duplicated(user_id, fromLast = TRUE))

## non-tax related [english] tweets sent to:trump
not_tax <- search_tweets(
  "to:realdonaldtrump -tax -taxes lang:en",
  n = 5000,
  include_rts = FALSE
)

## filter first tweet from each user
not_tax <- dplyr::filter(not_tax, !duplicated(user_id, fromLast = TRUE))

## add topic variable
yes_tax$topic <- "About Taxes"
not_tax$topic <- "Not Taxes"

## randomly sample from 'not_tax' for equal N and then bind rows
tt <- not_tax %>%
  dplyr::slice(sample(seq_len(nrow(not_tax)), nrow(yes_tax))) %>%
  list(yes_tax) %>%
  do.call(rbind, .)

## estimate sentiment
tt$sent <- syuzhet::get_sentiment(tt$text, method = "afinn")

## generate density-like plot (switch axes for horizontal-looking layout)
base_plot <- tt %>%
  ## sent scores are integers–so this adjusts values by topic to avoid overlap
  dplyr::mutate(sent = ifelse(
    topic == "Not Taxes", sent + .25, sent - .25
  )) %>%
  ggplot(aes(x = topic, y = sent, color = topic)) +
  ggbeeswarm::geom_beeswarm(alpha = .5) +
  scale_x_discrete(position = "top") +
  coord_flip()

## add labels and stylize plot
fancy_plot <- base_plot +
  labs(x = NULL, y = "Sentiment",
    title = "Sentiment of tweets to @realDonaldTrump by topic",
    subtitle = "Estimates for tweets calculated using the 'afinn' " %P%
      "sentiment dictionary") +
  dataviz::theme_mwk(base_size = 16) +
  theme(legend.position = "none") +
  scale_color_manual(
    values = c('Not Taxes' = "#1133ee", 'About Taxes' = "#dd2222"))

## save plot image to dropbox
fancy_plot + ggsave("~/Dropbox/sent-tax-tweets.png")

## compare means/standard deviations
tt %>%
  dplyr::group_by(topic) %>%
  dplyr::summarise(n = dplyr::n(),
    sent_mean = mean(sent, trim = .05),
    sent_sd = sd(sent))

When I ran this code on Feb 7th, 2019, the plot output looks like this: