sex_classify_friends.R

#CRAN versions dont work
#devtools::install_github("lmullen/gender")
#devtools::install_github("lmullen/genderdata")
library(rvest)
library(kirkegaard)
library(gender)
library(patchwork)

theme_set(theme_classic())

#read file
x = read_html("~/Documents/facebook-emilkirkegaard/friends/friends.html")

#get names
x %>%
  html_nodes("._2lel") %>%
  html_text() ->
  full_names

#first names
first_names = full_names %>% str_match("^[^ ]+") %>% as.vector()
#count doesn't match count on site, but it's close enough 326/339=96%

#try classify
gender::gender(first_names, method = "napp") -> results

results$proportion_male %>% describe()
results$proportion_male %>% round() %>% table2()

#try another method
gender::gender(first_names, method = "ssa") -> results2

results2$proportion_male %>% describe()
results2$proportion_male %>% round() %>% table2()

#visual representations
#split 5 ways
results$male = discretize(results$proportion_male, breaks = c(0, .1, .3, .7, .9, 1))
results2$male = discretize(results2$proportion_male, breaks = c(0, .1, .3, .7, .9, 1))
levels(results$male) = c("Female", "Probably female", "🤷", "Probably male", "Male")
levels(results2$male) = c("Female", "Probably female", "🤷", "Probably male", "Male")

results %>%
  ggplot(aes(male)) +
  geom_bar(aes(y = (..count..)/sum(..count..))) +
  scale_y_continuous("Percent", labels = scales::percent, breaks = seq(0, 1, .1)) +
  xlab("According to algorithm 1") +
  results2 %>%
  ggplot(aes(male)) +
  geom_bar(aes(y = (..count..)/sum(..count..))) +
  scale_y_continuous("Percent", labels = scales::percent, breaks = seq(0, 1, .1)) +
  xlab("According to algorithm 2")
GG_save("~/Pictures/facebook_friends_sex.png")