Who Mentions Who in the Office?

Jim: A Case Study

Who is talking to who in the Office?

Now that we have keep_characters, we can filter according to it and spit out who mentions who among the most relevant Office characters.

transcripts_tokenized %>% 
  filter(character %in% keep_characters) %>% 
  mutate(jim = ifelse(word == "jim", 1, 0)) %>% 
  group_by(character) %>% 
  summarise(jim = sum(jim)) %>% 
  arrange(desc(jim)) %>% 
  mutate(character = reorder(character, jim)) %>% 
  ggplot(ggplot2::aes(character, jim)) +
  geom_col() +
  coord_flip() +
  fix_bars() +
  labs(title = "Who Mentions Jim?",
       subtitle = "Counts of 'Jim' in The Office Transcripts",
       x = element_blank(),
       y = "Mentions")

The takeaway here is that Dwight mentions Jim the most, followed by Michael. No surprise there! What I find interesting is that only three characters really talk about/to Jim. After Dwight, Michael, and Pam (and Jim referencing himself, apparently), the mention rate for Jim’s name drops from over 200 to only 60 mentions. It seems as if the writers of the Office intentionally made Jim a subject of conversation among only a few characters!

Replicate for the rest of the cast

Next, we replicate that process for the rest of the cast. There is probably a better way to do this.

data_chord <- transcripts_tokenized %>% 
  filter(character %in% keep_characters) %>% 
  mutate(jim = ifelse(word == "jim", 1, 0)) %>% 
  mutate(michael = ifelse(word == "michael", 1, 0)) %>% 
  mutate(dwight = ifelse(word == "dwight", 1, 0)) %>% 
  mutate(pam = ifelse(word == "pam", 1, 0)) %>% 
  mutate(andy = ifelse(word == "andy", 1, 0)) %>% 
  mutate(angela = ifelse(word == "angela", 1, 0)) %>% 
  mutate(kevin = ifelse(word == "kevin", 1, 0)) %>% 
  mutate(erin = ifelse(word == "erin", 1, 0)) %>% 
  mutate(oscar = ifelse(word == "oscar", 1, 0)) %>% 
  # mutate(ryan = ifelse(word == "ryan", 1, 0)) %>% 
  # mutate(darryl = ifelse(word == "darryl", 1, 0)) %>% 
  # mutate(phyllis = ifelse(word == "phyllis", 1, 0)) %>% 
  # mutate(kelly = ifelse(word == "kelly", 1, 0)) %>% 
  # mutate(toby = ifelse(word == "toby", 1, 0)) %>% 
  group_by(character) %>% 
  summarise_at(vars(jim:oscar), funs(sum))

Make It Interactive

With nine people, some of the data can get easily concealed (how often did Angela mention Michael’s name?). One way to fix this is to make the visualization interactive, so that a user can hover over chords to see relationships between characters.

First, we conduct some data cleaning. I found that the rownames and column names have to be of the same order; let’s do a little manipulation to get there:

int_chord <- as.data.frame(data_chord)

rownames(int_chord) <- int_chord$character

row.order <- c("Jim", "Michael", "Dwight", "Pam", "Andy", "Angela", "Kevin", "Erin", "Oscar")
               #, "Ryan", "Darryl", "Phyllis", "Kelly", "Toby")
int_chord <- int_chord[row.order,]

Next, we load Matt Flor’s chorddiag package, and construct a matrix according to its function’s liking:

# devtools::install_github("mattflor/chorddiag")
library(chorddiag)

m <- as.matrix(int_chord[-1])

dimnames(m) <- list(have = int_chord$character,
                    prefer = str_to_title(colnames(int_chord[-1])))

Finally, we add a color palette and construct the diagram.

groupColors <- c("#B997C7", "#824D99", "#4E78C4", "#57A2AC", "#7EB875", "#D0B541", "#E67F33", "#CE2220", "#521A13")

p <- chorddiag(m, 
               groupColors = groupColors, 
               groupnamePadding = 35,
               tickInterval = 50,
               groupnameFontsize = 12)
p

# save the widget
# library(htmlwidgets)
# saveWidget(p, file="chord_interactive.html")

Play around with the diagram here!