trolls_neural_networks.R

library(tidyverse)
library(kerasformula)
library(textfeatures)

tweets <- read_csv("https://raw.githubusercontent.com/fivethirtyeight/russian-troll-tweets/master/IRAhandle_tweets_1.csv", 
                   n_max = 25000) %>% 
  mutate(text = content)

features <- textfeatures(tweets, threads = 2)

tweets_features <- bind_cols(tweets, features)

russian_trolls <- kms(account_category ~ 
                        following + followers + language + retweet + 
                        n_urls + n_hashtags + n_mentions + n_chars + n_commas + 
                        n_digits + n_exclaims + n_extraspaces + n_lowers + n_lowersp + 
                        n_periods + n_words + n_caps + n_nonasciis + n_puncts + 
                        n_capsp + n_charsperword + sent_afinn + sent_bing + 
                        n_polite + n_first_person + n_first_personp + 
                        n_second_person + n_second_personp + n_third_person + 
                        n_tobe + n_prepositions, 
                N_layers = 3, data = tweets_features, seed = 123)
russian_trolls$evaluations$acc
#> [1] 0.8706308

plot_confusion(russian_trolls)