Applying topic analysis of nzpol and relating it to user accounts

analysing_nzpol_themes.R

library(rtweet)
library(dplyr)
library(lubridate)
library(ggplot2)
library(tidytext)
library(topicmodels)
search_term <- "#nzpol"
corpus <- search_tweets(search_term, n=18000)
nzpol_utm <- corpus %>% select(screen_name, text) %>%
  unnest_tokens(word,text, token="tweets") %>%
  anti_join(stop_words) %>%
  group_by(screen_name) %>%
  count(screen_name, word) %>%
  cast_dtm(screen_name, word, n)
nzpol_lda <- LDA(nzpol_utm, k = 12, control = list(seed = 1234))
user_topics <- tidy(nzpol_lda, matrix = "beta")
top_terms <- user_topics %>% 
  group_by(topic) %>%
  slice_max(beta, n = 20) %>% 
  ungroup() %>% 
  arrange(topic, -beta)
# then just looked through top terms to form an opinion of what the
# topics were
users_gamma <- tidy(nzpol_lda, matrix = "gamma")
user_focus <- users_gamma %>%
  group_by(document) %>%
  mutate(focus = gamma - mean(gamma)) %>%
  arrange(desc(focus)) %>%
  slice(1) %>%
  ungroup() %>%
  rename(screen_name=document)
nzpol_focus <- corpus %>% 
  group_by(screen_name, account_created_at) %>%
  summarise(account_age = as.numeric(difftime(Sys.time(), account_created_at[1], units="mins"))) %>% ungroup() %>%
  inner_join(user_focus) %>%
  mutate(age_ntile = ntile(account_age, 20),
         age_group = ifelse(age_ntile == 1, "newest_0.05", "older accounts")) %>%
  filter(ntile(focus, 20) == 20) %>%
  count(topic, age_group) %>%
  group_by(age_group) %>%
  mutate(percentage = n/sum(n)) %>%
  ungroup()

ggplot(nzpol_focus, aes(x=factor(topic), y=percentage, fill=age_group)) + 
  geom_col(position=position_dodge()) + theme_minimal()