How Exciting: Algorithm Detects Sarcasm

The amusing nerds over at Geekosystem, and the more serious nerds over at Slashdot are reporting in that someone has created an algorithm capable of picking sarcasm in written statements. Great, just great. That immediately ruins the chance of ever messing with some straight-laced, sensible robot slave in the future.

It’s called SASI (semi-supervised sarcasm identification algorithm), and apparently “SASI achieved a precision of 77% and recall of 83.1% “on an evaluation set containing newly discovered sarcastic sentences, where each sentence was annotated by three human readers.”” More info and a couple of thoughts after the jump.

It was tested on Amazon comments/reviews and Tweets, and while the Internet is abundant with sarcasm, I really feel they should have forced it to converse with teenagers. I wonder how it would cope with The Onion?

Report: 70 Percent Of All Praise Sarcastic

Regardless, this could have some interesting implications for the growing social listening/monitoring industry. There are plenty of social monitoring tools that offer sentiment measurement, but to be honest, none of the automatic tools do it terribly well – and the measurement solutions that use teams of outsourced workers aren’t exactly cost effective.

If an algorithm can deftly pick sarcasm, something a good portion of the human population has trouble with, then surely the other emotions out there shouldn’t be too hard.


Like This!

Add to FacebookAdd to DiggAdd to Del.icio.usAdd to StumbleuponAdd to RedditAdd to BlinklistAdd to TwitterAdd to TechnoratiAdd to Yahoo BuzzAdd to Newsvine

Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: