Whereas any big information website quotes at this very moment the impressive figures of the repercussions of the American election on Twitter in term of number of micro-messages, Pegasus offers a targeted and qualitatively instructive message recovery.
This study retrieves all messages containing both the name of a candidate ("Obama" and "Romney") and an emoticon affixed to it, this last one being either a positive or a negative smiley.
After all, no matter if the smiley is positive or not, this is only an indicator of a personal message!
Basically, if we do not offer an analysis of the total corpus of millions of tweets sent during these presidential election days, it’s because we prefer to focus on personalised messages, vectors of an emotional information, rather than on the many "retweets" posted by various agencies, medias, campaigns’ committees and professional / amateurs analysts!
Our infographics : a sample of 200’000 tweets
The graphics below replaces the hundreds of thousands of posts on a vertical time axis. On the right, the reader finds a histogram that allows to realise and understand the tweets’ volume issued to be able to establish the central table.
We observe that in the plebiscite of Barack Obama precede a period during which the mentions of Mitt Romney are more numerous than the average. It’s about the effect of the announcement of the first results, for the greater part favorable to the republican candidate.
To sum up, a situation globally very favorable to the outgoing president:
What relevance for a more "qualitative" selection ?
The method we use is rather simple. By going through the basic smileys ( ":)", ":-)", ":(" et ":-(" ) and associating them with the names of both main candidates in our keyword searches, we are able to retrieve more personalised messages, and widely distributed in the population on Twitter, unlike tweets of newspapers and other teams of campaign monopolizing attention.
This way of working has been used and studied by [Pak & Paroubek, 2010] with significant results. Twitter also offers this service: while writing a smiley in the search field in addition to the words defining the targeted subject, you will be able to see all tweets corresponding to the selected smiley alongside with your subject’s search. We can deduce that in the same tweet several contradictory smileys can appear, or that the names of both candidates appear to it. In both cases, these results just nullify themself during the counting; we don’t need to worry about it.
To pursue the study of this corpus the words chosen in the text could be analyzed. Words as " like ", "love", "yes" or " lose ", and "sad" allow to seize the feelings of the author, quite as to balance them ("yes" is weaker than "YEEEEEEEEEEEEEES " for example).
Interpreting the results
The smiley, a problem
At last, it’s obvious that the use of a name of a candidate and a smiley won’t surely enough determine the intention of the writer. An example of tweet which corresponds to the intuitive interpretation:
Now that Obama won, I might actually move to Chile :’(
— Ariadna (@AriP_xx) Novembre 7, 2012
An example of tweet that will mislead this approach:
— Heather H. (@HHartwellz) Novembre 7, 2012
Let’s put in perspective this bias by two observations:
Confronted to an important mass of data, and after a random perusal, it seems that the "non-intuitive" (for example, one delighted at the defeat of Obama in a State, thus including "Obama ") is flooded in the mass of intuitive tweets (a tweet which is delighted at the victory of Obama in a State, thus including "Obama ").
- Our objective in this research is not to interpret these results by pulling a report as " Obama is the most appreciated " but to confine our corpus of analysis in a panel qualitatively more interesting than the totality of the emitted messages. Therefore, this is a way to take into account only a sample of more personal messages than the majority of tweets drafted during those few dozen hours.