On January 5th 2017, the Obama administration relased their social media data.

Among these data were all of the tweets the former president made from the @POTUS account.

Dan Shiffman cleaned up the text format slightly for the Obamathon event at ITP - NYU. This cleaned data - found here - forms the basis for this project.

For the sentiment analysis, the AFINN-111 data set is used. It contains 2477 words which has been manually rated on a scale from -5 to 5.

Regular expressions are used to find the word in the AFINN data and return sentiment:

sentiment <- function(word) {
  lookup <- grep(paste0("^", word, "$"), afinn$word)
  if (length(lookup) == 0) {
    return(0)
  }
  afinn[lookup,]$score[1]
}

sapply(c("joy", "wrong", "dachshund"), sentiment)

##       joy     wrong dachshund 
##         3        -2         0

We want to use CSS for coloring words with non-zero sentiment:

insert_sentiment_span <- function(string, word, sentiment) {
  capture <- paste0("(", word, ")")
  replacement <-
    paste0('<span class="sentiment',
           as.character(sentiment),
           '">\\1</span>')
  gsub(capture, replacement, string, ignore.case = TRUE)
}

insert_sentiment_span("Happy birthday!", "happy", 3)

## [1] "<span class=\"sentiment3\">Happy</span> birthday!"

This process is repeated to get a sentiment score for the tweet. It is then compared to the overall distribution of sentiments, and the average sentiment at the time of the tweet.