resist_oped
🕵🏽 ♀️ Identifying the author behind New York Time’s op-ed from inside the Trump White House.
## corr matrix
mat <- readRDS("data/mat.rds")
## print matrix
cor(t(mat))[, "op-ed"] %>% sort()
#> USDOT DeptVetAffairs USDOL HUDgov
#> -0.7923443 -0.6046442 -0.5116041 -0.4989040
#> EPA USDA Interior SBAgov
#> -0.4929554 -0.4810160 -0.4326364 -0.3840941
#> USTreasury SecretaryZinke SecShulkin DHSgov
#> -0.3413849 -0.2922546 -0.2605447 -0.2237110
#> TheJusticeDept SecPriceMD Cabinet usedgov
#> -0.2231511 -0.2102655 -0.1861103 -0.1610005
#> SecretaryPerry SecElaineChao HHSGov ENERGY
#> -0.0998643 -0.0936370 -0.0794056 -0.0606396
#> mike_pence OMBPress CommerceGov nikkihaley
#> -0.0563475 -0.0364755 -0.0270086 0.0452229
#> SBALinda CIA DeptofDefense USUN
#> 0.1009770 0.1139948 0.1398299 0.1790831
#> USTradeRep SecretaryAcosta stevenmnuchin1 SecretarySonny
#> 0.1840837 0.1860960 0.1905300 0.3356500
#> SecretaryCarson EPAAWheeler BetsyDeVosED SecAzar
#> 0.4894218 0.4931722 0.5488558 0.6268701
#> SecNielsen StateDept SecretaryRoss SecPompeo
#> 0.6716064 0.6843869 0.7373439 0.7669710
#> POTUS VP op-ed
#> 0.7986615 0.8010629 1.0000000
## plot estimates
est <- cor(t(mat))[, "op-ed"] %>% sort()data_frame(screen_name = names(est), r = est) %>%
arrange(est) %>%
mutate(screen_name = factor(screen_name, levels = screen_name)) %>%
ggplot(aes(x = screen_name, y = r)) +
geom_col(aes(fill = r > 0), width = .3) +
geom_point(shape = 21, fill = "greenyellow", size = 2.75) +
tfse::theme_mwk(light = "white", base_family = "Avenir Next LT Pro") +
coord_flip() +
labs(title = "Text similarity of NYT op-ed and Cabinet tweets",
subtitle = "Correlation estimates based on 100+ features extracted from texts",
y = NULL, x = NULL,
caption = "Source: Texts from Twitter & New York Times analyzed by @kearneymw") +
theme(legend.position = "none") +
scale_fill_manual(values = c("#2244ee", "#dd2222")) +
ggsave("plot.png", width = 7, height = 6.5)
</>
Data
I compared the paragraphs of the op-ed to tweets posted to timelines by members of the Cabinet.
Code
Here’s the code:
## load tidyverse
library(tidyverse)
library(rtweet)
library(rvest)
## download Twitter profiles via CSPAN's cabinet list
cab_twits <- lists_members(
owner_user = "CSPAN", slug = "the-cabinet")
## get up to 3200 of most recent tweets for each
cab_tweets <- cab_twits %>%
filter(screen_name != "realDonaldTrump") %>%
pull(user_id) %>%
map(get_timeline, n = 3200) %>%
bind_rows()
## scrape source code for op-ed
nyt <- read_html(
"https://www.nytimes.com/2018/09/05/opinion/trump-white-house-anonymous-resistance.html")
## return just the paragraph text
nyt_text <- nyt %>%
html_nodes("p") %>%
html_text() %>%
.[3:31]
## create data set with just author (id) and text
data <- data_frame(
id = c(cab_tweets$screen_name, rep("op-ed", length(nyt_text))),
text = c(cab_tweets$text, nyt_text)
)
## feature extraction
tf <- textfeatures::textfeatures(data, word_dims = 80, threads = 20)
## summarise by id
tfsum <- tf %>%
group_by(id) %>%
summarise_all(mean, na.rm = TRUE) %>%
ungroup()
## vector of unique authors
authors <- unique(tfsum$id)
## create numeric vectors of equal length for each author
cols <- map(authors,
~ filter(tfsum, id == .x) %>% select(-id) %>% as.list() %>% unlist())
## create matrix
mat <- cols %>%
unlist() %>%
as.numeric() %>%
matrix(nrow = length(authors), byrow = TRUE)
## set row and column names
row.names(mat) <- authors
## dipslay matrix
cor(t(mat))[, "op-ed"] %>% sort()