little_women.R

library(tidyverse)
library(tidytext)
library(gutenbergr)
library(cleanNLP)

little_women_raw <- gutenberg_download(514, meta_fields = "title")

little_women <- little_women_raw %>% 
  slice(70:n()) %>% 
  mutate(chapter_start = str_detect(text, "^CHAPTER"),
         chapter_number = cumsum(chapter_start)) %>% 
  select(-gutenberg_id, -title, -chapter_start)

# Structure the data for the POS tagger
little_women_to_tag <- little_women %>% 
  group_by(chapter_number) %>% 
  nest(data = c(text)) %>% 
  ungroup() %>% 
  mutate(text = map_chr(data, ~paste(.$text, collapse = " "))) %>% 
  select(-data)

# Use the R-only part-of-speech tagger
cnlp_init_udpipe()

# This took ≈180 seconds to run
little_women_tagged <- cnlp_annotate(little_women_to_tag, 
                                     text_name = "text", 
                                     doc_name = "chapter_number")
little_women_tokens <- little_women_tagged$token

proper_nouns <- little_women_tokens %>% 
  filter(upos == "PROPN")

main_characters_by_chapter <- proper_nouns %>% 
  filter(lemma %in% c("Meg", "Jo", "Beth", "Amy")) %>% 
  group_by(doc_id, lemma) %>% 
  summarize(n = n()) %>% 
  mutate(name = factor(lemma, levels = c("Meg", "Jo", "Beth", "Amy"), ordered = TRUE)) %>% 
  rename(chapter = doc_id) %>% 
  group_by(chapter) %>% 
  mutate(prop = n / sum(n)) %>% 
  ungroup() %>% 
  mutate(chapter_name = paste("Chapter", chapter)) %>% 
  mutate(chapter_name = fct_inorder(chapter_name))

# Meh
ggplot(main_characters_by_chapter, aes(x = name, y = n, fill = name)) + 
  geom_col() +
  facet_wrap(vars(chapter))

# Super meh
ggplot(main_characters_by_chapter, aes(x = name, y = 1, size = n, color = name)) + 
  geom_point() +
  facet_wrap(vars(chapter))

# Awesome
ggplot(main_characters_by_chapter, aes(x = prop, y = 1, fill = fct_rev(name))) + 
  geom_col(position = position_stack()) +
  scale_x_continuous(expand = c(0, 0)) +
  scale_y_continuous(expand = c(0, 0)) +
  scale_fill_viridis_d(option = "plasma", end = 0.9, name = NULL) +
  guides(fill = guide_legend(reverse = TRUE)) +
  labs(x = NULL, y = NULL,
       title = "Proportion of mentions of each\nLittle Woman per chapter",
       subtitle = "Jo basically dominates the last third of the book") +
  facet_wrap(vars(chapter_name), nrow = 6) +
  theme_bw(base_family = "Halis GR") +
  theme(legend.position = "top",
        axis.text = element_blank(),
        axis.ticks = element_blank(),
        strip.background = element_rect(fill = "white"),
        legend.text = element_text(face = "bold", size = rel(1)),
        plot.title = element_text(face = "bold", hjust = 0.5, size = rel(1.7)),
        plot.subtitle = element_text(hjust = 0.5, size = rel(1.1)))