What

This is a package that does/has only one thing: the complete transcriptions of all episodes of The Office! (US version).

Use this data set to master NLP or text analysis. Let’s scratch the surface of the subject with a few examples from the excellent Text Mining with R book, by Julia Silge and David Robinson.

First install the package from CRAN:

library(schrute)

There is only one data set with the schrute package; assign it to a variable

mydata <- schrute::theoffice

Take a peek at the format:

dplyr::glimpse(mydata)
#> Observations: 55,130
#> Variables: 9
#> $ index            <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, …
#> $ season           <chr> "01", "01", "01", "01", "01", "01", "01", "01",…
#> $ episode          <chr> "01", "01", "01", "01", "01", "01", "01", "01",…
#> $ episode_name     <chr> "Pilot", "Pilot", "Pilot", "Pilot", "Pilot", "P…
#> $ director         <chr> "Ken Kwapis", "Ken Kwapis", "Ken Kwapis", "Ken …
#> $ writer           <chr> "Ricky Gervais;Stephen Merchant;Greg Daniels", …
#> $ character        <chr> "Michael", "Jim", "Michael", "Jim", "Michael", …
#> $ text             <chr> "All right Jim. Your quarterlies look very good…
#> $ text_w_direction <chr> "All right Jim. Your quarterlies look very good…

 mydata %>%
  dplyr::filter(season == '01') %>%
  dplyr::filter(episode == '01') %>%
  dplyr::slice(1:3) %>%
  knitr::kable()

So what we have is the season, episode number and name, character, the line spoken and the line spoken with the stage direction (cue).

We can tokenize all of the lines with a few lines from the tidytext package:

token.mydata <- mydata %>%
  tidytext::unnest_tokens(word, text)

This increases our data set to 570566 records, where each record contains a word from the script.

 token.mydata %>%
  dplyr::filter(season == '01') %>%
  dplyr::filter(episode == '01') %>%
  dplyr::slice(1:3) %>%
  knitr::kable()

If we want to analyze the entire data set, we need to remove some stop words first:

stop_words <- tidytext::stop_words

tidy.token.mydata <- token.mydata %>%
  dplyr::anti_join(stop_words, by = "word")

And then see what the most common words are:

tidy.token.mydata %>%
  dplyr::count(word, sort = TRUE) 
#> # A tibble: 19,041 x 2
#>    word        n
#>    <chr>   <int>
#>  1 yeah     2928
#>  2 hey      2231
#>  3 michael  1859
#>  4 uh       1458
#>  5 gonna    1398
#>  6 dwight   1339
#>  7 jim      1168
#>  8 time     1146
#>  9 pam      1044
#> 10 guys      945
#> # … with 19,031 more rows

tidy.token.mydata %>%
  dplyr::count(word, sort = TRUE) %>%
  dplyr::filter(n > 400) %>%
  dplyr::mutate(word = stats::reorder(word, n)) %>%
  ggplot2::ggplot(ggplot2::aes(word, n)) +
  ggplot2::geom_col() +
  ggplot2::xlab(NULL) +
  ggplot2::coord_flip() +
  ggplot2::theme_minimal()

Feel free to keep going with this. Now that you have the time line (episode, season) and the character for each line and word in the series, you can perform an unlimited number of analyses. Some ideas: - Sentiment by character - Sentiment by character by season - Narcissism by season (ahem.. Nard Dog season 8-9) - Lines by character - Etc.