Video Game Title Emotions - a text analysis

Using R to create a data visualization of the emotional content of video game titles

Figure 1: Video Game Title Emotions

Video Game Title Emotions - a text analysis

This week’s Tidy Tuesday challenge offered video game data for analysis. I analyzed the text in the video game titles. Using the NRC Emotion Lexicon I measured the joy words and sad words in the titles by year.

I used the tidytext package to tokenize each years-worth of game titles and remove stop words and plotted the polar plots with the gpglot2 package. Then I made word clouds using wordcloud2. The clowplot package was used to place screenshots of the wordclouds within the final image.

Code is also on GitHub

library(wordcloud2)
library(cowplot)
library(tidyverse)
library(tidytext)


get_sentiments("nrc")

data(stop_words)

nrc_types <- get_sentiments("nrc") %>% 
 pull(sentiment) %>% 
 unique()

nrc_joy <- get_sentiments("nrc") %>% 
 filter(sentiment == "joy")
nrc_sad <- get_sentiments("nrc") %>% 
 filter(sentiment == "sadness")

stops <- stop_words %>% 
 bind_rows(tibble(
  word = c("edition", "online", "ii", "iii", "iv", "vi", "vii",
           "viii", "ix", "definitive", "remastered"),
  lexicon = c(rep("jda", 11))
 ))

games <- readr::read_csv('https://raw.githubusercontent.com/rfordatascience/tidytuesday/master/data/2021/2021-03-16/games.csv')

# function to get emotion words and output as list of two types
get_emotion_words <- function(x) {
 
 name <- as_tibble(x) %>% 
  rename(gamename = value) %>% 
  count(gamename) %>% 
  mutate(gamename = str_remove_all(gamename, "<U.....>")) %>% 
  select(-n) %>%
  unnest_tokens(
   output = word,
   input = gamename,
   token = "words",
   to_lower = TRUE,
   strip_punct = TRUE
  ) %>%
  anti_join(stops) %>% 
  filter(
   nchar(word) > 1,              # must be more than one character
   str_detect(word, "^[a-z]"),   # must start with a letter
   str_detect(word, "[a-z]")     # must not be only numbers
  ) %>% 
  count(word, sort = TRUE) %>%
  mutate(word = reorder(word, n))
 
 emotion_words <- list(
  joy = name %>% inner_join(nrc_joy),
  sad = name %>% inner_join(nrc_sad)
 )
 
 emotion_words
 
}

name_emotions <- games %>%
 filter(year > 2014 & year < 2021) %>% 
 group_by(year) %>% 
 summarise(sum_avg_players = sum(avg),
           emotion_words = get_emotion_words(gamename)) %>% 
 ungroup() %>%
 arrange(year)

emotions_scored <- name_emotions %>% 
 unnest(emotion_words) %>% 
 mutate(score = n*sum_avg_players) %>% 
 group_by(year, sentiment) %>% 
 summarise(valence = sum(score)) %>% 
 ungroup() %>% 
 mutate(sentiment = factor(sentiment, levels = c("sadness", "joy"), order = TRUE))

joy_title_words <- name_emotions %>%
 unnest(emotion_words) %>%
 select(word, sentiment, freq = n) %>% 
 filter(sentiment == "joy") %>% 
 select(-sentiment) %>%
 group_by(word) %>% 
 summarise(freq = sum(freq))

sad_title_words <- name_emotions %>%
 unnest(emotion_words) %>%
 select(word, sentiment, freq = n) %>% 
 filter(sentiment == "sadness") %>%
 select(-sentiment) %>%
 group_by(word) %>% 
 summarise(freq = sum(freq))

# open these in the viewer pane then screenshot and add screenshot to directory
joy_cloud <- wordcloud2(
 data = joy_title_words,
 size = .7,
 gridSize = 1,
 color = "#e600e6",
 backgroundColor = "black",
 minRotation = 0,
 maxRotation = 0,
 rotateRatio = 0,
 shape = "circle"
)

sad_cloud <- wordcloud2(
 data = sad_title_words,
 size = 1.1,
 gridSize = 1,
 color = "#625062",
 backgroundColor = "black",
 minRotation = 0,
 maxRotation = 0,
 rotateRatio = 0,
 shape = "circle"
)

p <- emotions_scored %>%
 ggplot(aes(x = sentiment, y = valence, fill = sentiment)) +
 geom_bar(stat = "identity", width = 1) +
 scale_fill_manual(values = c("#e600e6", "#625062"), breaks = c("joy", "sadness")) +
 labs(x = "",
      y = "",
      title = "Emotional Content of Words in Video Game Titles
In all years, game titles consisted of more sad words than joy words.",
      subtitle = "Emotional valence of words determined by the NRC Emotion Lexicon",
      caption = "Data source: SteamCharts | Data viz: @jeremy_data
Reference: Emotions Evoked by Common Words and Phrases: Using Mechanical Turk to Create an Emotion Lexicon, Saif Mohammad and Peter Turney, In Proceedings of the NAACL-HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, June 2010, LA, California.",
      fill = "") +
 coord_polar() +
 facet_wrap(~ year, ncol = 6) +
 theme_minimal() +
 theme(
  text = element_text(color = "#cfafcf"),
  strip.text = element_text(size = 12, color = "#cfafcf"),
  plot.title = element_text(size = 34, margin = margin(0,0,.25,0, unit = "cm")),
  plot.subtitle = element_text(size = 24, margin = margin(.5,0,4,0, unit = "cm")),
  plot.caption = element_text(size = 14, margin = margin(24,0,0,0, unit = "cm")),
  legend.box.margin = margin(.25,0,2,0, unit = "cm"),
  legend.position = "top",
  legend.text = element_text(size = 16),
  legend.spacing.x = unit(1, 'cm'),
  legend.text.align = 0,
  panel.grid = element_blank(),
  axis.title.x = element_blank(),
  axis.title.y = element_blank(),
  axis.text.x = element_blank(),
  axis.text.y = element_blank(),
  plot.margin = margin(3,3,.5,3, unit = "cm"),
  plot.background = element_rect(fill = "black"),
 ) 

q <- ggdraw(p) +
 draw_image(here::here("joycloud.png"), x = .125, y = .265, hjust = 0, vjust = 0, width = 0.25, height = 0.15) +
 draw_image(here::here("sadcloud.png"), x = .48, y = .12, hjust = 0, vjust = 0, width = 0.37, height = 0.37)

ggsave("sad_games.png", plot = q, width = 30, height = 26, units = "in")