Could sentiment analysis — the automated mining of attitudes, opinions, and emotions from text, speech, and database sources — have foretold the demise of Egyptian autocrat Hosni Mubarak?
Can this fast-emerging technology and discipline predict the movement of Oracle’s share price based on online and social reactions to company and market news? Could it quantify reactions to Groupon’s widely panned SuperBowl ad and tell the ad-agency creative types what particular aspects viewers disliked? And in a more mundane, everyday application, could it identify product defects to help convert dissatisfied customers into promoters?
These are four of seven sentiment-analysis scenarios that I’ll describe, ways the technology can be (and is being) applied to derive insight from qualitative information sources.
This is a really cool idea and would have tremendous value in finding insights behind mounds of data. I am a little skeptical of the accuracy of sentiment analysis. If people sometimes misinterpret the tone and voice of writing, I’m not sure about software’s ability to do so. Regardless, this is a new term to me that I’m interested to learn more about.
Thanks for the comment, Kevin. The short answer on accuracy is: 1) Humans aren’t perfect. The most scientific study I’ve seen found only 81% agreement in sentiment classification by 2 people, boosted to 90% when cases they agreed were uncertain were removed. 2) Accuracy can be improved by hybrid human-machine systems. Use the machines for what they’re good at — speed, capacity, ability to work in multiple languages, ability to organize — guide by human judgment. 3) Accuracy needs should be evaluated not on some absolute scale but relative to the business (or government) problem you’re trying to solve.
I’ll put in a plug: To learn more, consider attending a conference I’m organizing, the Sentiment Analysis Symposium, April 12 in New York, http://sentimentsymposium.com. It’s preceded on April 11 by a half-day Practical Sentiment Analysis tutorial to be taught by staff from eBay.