I recently had the privilege of reviewing internal award nominations. In a secret duel of persuasion, hundreds of employees seized the opportunity to outboast their colleagues in boasting about their colleagues. Letterhead was prepared, anecdotes collected, and thesaureses consulted.1
While the review process on the whole was a charming experience, the sheer volume of submissions made the work somewhat tedious. The word count ran into the tens of thousands. Each nomination required a human’s careful perusal, but it wasn’t long before I was imagining a more programmatic approach. That meant R, and that meant
For reference, each submission looked like this:
Your First Name: Lorum ipsum
Your Last Name: Lorum ipsum
Your Email: Lorum ipsum
Nominee’s First Name: Lorum ipsum
Nominee’s Last Name: Lorum ipsum
Nomination Statement: Lorum ipsum blah blah
My plan was this: read the folders of PDFs into R, scrape and tidy the text, and calculate a winner based on a formula that weighs total nominations, total words, and total positive words. The nominations came in a zip file, with sub-directories of each nominee, and then PDFs of the nomination submissions within the sub-directories. I began by unzipping the PDFs and ingesting the text:
The relevant text occurs after “Statement:”, so I was only interested in the text after that point.
nom_names is derived from
the file paths.
Now the show begins. Below I: (1) build the
nominations_df data frame; (2) calculate the number of nominations per employee,
(3) unnest the words; (4) total the words, positive words, and calculate the positive word ratio and nomination score; and
(5) order the data frame by nomination score. When this runs, I’ll see our winner!
I wanted the nomination score formula to weigh number of submissions more heavily than total words. For example, I am more impressed by three people writing 100 words each than I am by one person writing 300 words. And while fraught with ambiguities, I like having the positive word ratio as a coefficient of total words. Thus, this formula rewards nominees with multiple nominators who both wrote a lot and wrote positively.
There are other, more interesting analyses to run: which positive words occur most? Is there an observable difference in the positive words used in the nominations of male vs. female employees? The data is rich, but very much private. The above is just a playful, albeit lazy idea.
But far be it from me to deny the nominators and selection committee my full subjectivity and whatever measure of arbitrariness included. Much better to read the submissions myself.
I’m told the plural of “thesaurus” can also be “thesauri”. ↩