By Kevin Kline
In a few days, American voters will decide who be the next President of the United States. Most predictions have former Secretary of State Hillary Clinton with a slightly higher chance of winning the election than Republican candidate Donald Trump.
Of course, this election process has been SO typical and predictable.
That was sarcasm. While you probably caught onto it, computers have a hard time detecting sarcasm and other instances of “figurative language.” University College Dublin postgrad student Aniruddha Ghosh’s Twitter bot called Sarcasm Magnet is getting better at detecting sarcasm. The bot constantly learns from feedback on Twitter.
The bot isn’t perfect. It looks for keywords and context but isn’t fully human. However, the feedback taught Ghosh something: sarcasm is a great tool to calculate campaign support.
Ghosh found when Hillary Clinton spoke in traditionally Republican-leaning communities, sarcasm would jump. When Trump spoke in Democratic cities, sarcasm would jump. There was an inverse connection between sarcasm and voter enthusiasm. As a side project to his broader research, Ghosh added a “social stance” function to the bot.
“It was just [something] for fun that I added,” said Ghosh. “I just want to see who wins if my bot can predict that.”
The side project involved doing sentiment analysis of his data. The presidential election lends itself perfectly to test out the power of the bot.
Ghosh may face a steep challenge. Firstly, American elections aren’t decided by popular vote outright. Instead it is state-by-state challenges. In most cases, those states are solidly Republican or Democrat. Only a handful are really toss ups.
Also, up until now, public opinion polling was a fairly strong indicator of how elections will end up. There are certainly challenges to the polling world like dropping response rates and fewer polls, but how does this student’s side project compare? Can it function better than actual surveys?
The way I see it after covering multiple American elections, Ghosh has a 1 in 4 chance of proving his bot’s powers. If he picks Trump and Trump loses, he’s wrong. The same goes if he picks Clinton and she loses. However, if he picks Clinton and she wins, can he really claim a victory? The Democratic candidate has an edge in 9 of the 10 most hotly contested state races. The bot simply corroborates what polls already suggested.
If Ghosh picks Trump and Trump wins, there could be something there. Ghosh plans to make his final prediction on Monday. We’ll find out whether he is right or wrong sometime Wednesday morning.