Text Analysis

Text analysis of Good Omens by Neil Gaiman and Terry Pratchett.

Armageddon Only Happens Once: Good Omens Text Analysis

Synopsis: The world is going to end next Saturday, just before dinner, but it turns out there are a few problems — the Antichrist has been misplaced, the Four Horseman of the Apocalypse ride motorcycles and the representatives from heaven and hell decide that they like the human race. (Synopsis from NPR)

Word Counts Visualizations

Figure 1. Word Count Plot

hide

#---------------------------------------
# Text Wrangling
#---------------------------------------

# Read in Good Omens pdf
omens_text <- pdf_text(here("_posts/2021-02-25-text-analysis","Media","GoodOmens.pdf"))

# Tidy
omens_tidy <- data.frame(omens_text) %>% 
  mutate(text_full = str_split(omens_text, pattern = "\\n")) %>% 
  unnest(text_full) %>% 
  mutate(text_full = str_trim(text_full)) %>% 
  slice(-(63:109)) %>% # Skip character descriptions
 mutate(chapter = case_when(
    str_detect(text_full, "^In the beginning$|^Eleven years ago$|^Wednesday$|^Thursday$|^Friday$|^Saturday$|^Sunday$") ~ text_full,
    TRUE ~ NA_character_
  )) %>% 
  fill(chapter) %>% 
  mutate(chapter = fct_relevel(chapter, 
            "In the beginning", "Eleven years ago", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday")) %>%  # Set chapter order
  unnest_tokens(word, text_full) %>% # Tokenize
  mutate(word = str_replace(word, "[0-9-]+", NA_character_)) %>% # remove rogue page numbers
  select(-omens_text) %>% 
  drop_na() %>% 
  anti_join(stop_words) # Remove stop-words

# Get word counts of non stop-words
nonstop_counts <- omens_tidy %>% 
  count(chapter, word)

#---------------------------------------
# Visualize top word counts by chapter
#---------------------------------------

# Find top 5 words by chapter
top_5_words <- nonstop_counts %>% 
  group_by(chapter) %>% 
  arrange(-n) %>% 
  slice(1:5)

# Visualize top 5 words per chapter

omens_palette <- c("#A6D3DD", "#FDF3B0", "#F0B140", "#B4B4B4", "#C6333B", "#671D21", "#070707")

ggplot(data = top_5_words, aes(x = word, y = n)) +
  geom_col(aes(fill = chapter), show.legend = FALSE) +
  scale_fill_manual(values = omens_palette) +
  facet_wrap(~chapter, scales = "free") +
  theme_minimal() +
  coord_flip() +
  labs(title = "Five most common words per chapter in Good Omens",
       x = "Word",
       y = "Number of occurences") +
    theme(plot.title = element_text(color="#C6333B", 
                                  size=16,
                                  face = "bold",
                                  hjust = 0.5))

Figure 2. Word Cloud Plot

hide

#---------------------------------------
# Visualize top 100 words in book
#---------------------------------------

omens_top100 <- nonstop_counts %>% 
  arrange(-n) %>% 
  slice(1:100) %>% 
  mutate(angle = 90 * sample(c(0, 1), n(), replace = TRUE, prob = c(60, 40)))

omens_cloud <- ggplot(data = omens_top100, aes(label = word)) +
  geom_text_wordcloud(aes(color = n, 
                          size = n,
                          angle = angle),
                          shape = "squares") +
  scale_size_area(max_size = 10) +
  scale_color_gradientn(colors = c(low = "#A6D3DD", middle = "#F0B140", high = "#C6333B")) +
  labs(title = "Top 100 Words in Good Omens") +
  theme(plot.title = element_text(color="#C6333B", 
                                  size = 16,
                                  face = "bold",
                                  hjust = 0.5))

# Add background
img = here("_posts/2021-02-25-text-analysis","Media","angel-demon-buds.png")

ggbackground(omens_cloud, img)

Sentiment Analysis

Figure 3. Positive vs. Negative Sentiment Analysis

hide

#---------------------------------------
# Sentiment analysis with afinn
#---------------------------------------

# Join afinn lexicon
omens_afinn <- omens_tidy %>% 
  inner_join(get_sentiments("afinn"))

# Calculate mean score per chapter
afinn_means <- omens_afinn %>% 
  group_by(chapter) %>% 
  summarize(mean_afinn = mean(value))

# Visualize positive vs. negative sentiment per chapter
ggplot(data = afinn_means, 
       aes(x = fct_rev(as.factor(chapter)), 
           y = mean_afinn)) +
  geom_col(aes(fill = chapter), show.legend = FALSE) +
  scale_fill_manual(values = omens_palette) +
  theme_minimal() +
  labs(
    title = "Positive and negative sentiments by chapter\n of of Good Omens",
    x = "Chapter",
    y = "Average Sentiment"
  ) +
  theme(plot.title = element_text(color="#C6333B", 
                                  size = 16,
                                  face = "bold",
                                  hjust = 0.5)) +
  coord_flip()

Figure 4. Sentiment Bins Plot

hide

#---------------------------------------
# Sentiment analysis with nrc
#---------------------------------------

# join nrc lexicon
omens_nrc <- omens_tidy %>% 
  inner_join(get_sentiments("nrc"))

# Get counts
omens_nrc_counts <- omens_nrc %>% 
  count(chapter, sentiment)

# Visualize sentiment bins by chapter
ggplot(data = omens_nrc_counts, aes(x = sentiment, y = n)) +
  geom_col(aes(fill = chapter), show.legend = FALSE) +
  scale_fill_manual(values = omens_palette) +
  theme_minimal() +
  labs(
    title = "Top sentiment bins by chapter of Good Omens",
    x = "Chapter",
    y = "Number of occurences"
  ) +
  theme(plot.title = element_text(color="#C6333B", 
                                  size = 16,
                                  face = "bold",
                                  hjust = 0.5)) +
  facet_wrap(~chapter) +
  coord_flip()

---

References:
Gaiman, N., & Pratchett, T. (1990). Good Omens: The Nice and Accurate Prophecies of Agnes Nutter, Witch. London: Berkley Books.

Mohammad, S. & Turney, P. (3013). Crowdsourcing a Word-Emotion Association Lexicon. Computational Intelligence, 29 (3), 436-465.

Nielsen, F.A. (2011). A new ANEW: Evaluation of a word list for sentiment analysis in microblogs. Proceedings of the ESWC2011 Workshop on ‘Making Sense of Microposts’: Big things come in small packages 718 in CEUR Workshop Proceedings 93-98.