• E-mail
• License
• Print
• Comment
• RSS

Twitter, Baby, Tell Me How You REALLY Feel

Dumenco's Trendrr Chart of the Week

by Simon Dumenco
Published: October 23, 2009

Since my next Media Guy column (in print Monday, and available online this Sunday evening) is titled "Top 10 Lessons to Learn From NBC's Failing Leno Strategy," I've been particularly curious about Jay Leno's Twitter buzz. As it happens, the folks at Trendrr -- the social-media tracking service -- just introduced something called Twitter sentiment analysis. The idea is that, by mapping certain words against the feelings they are typically meant to convey, you can get a sense of not only how many people are tweeting about a certain subject, but how they generally feel about it.

It is not, obviously, an exact science. Plenty of statisticians and tech heads around the world are trying to figure out the best way to parse the meaning of the zillions of snippets of text floating around in the social-media space (Trendrr's algorithm, FYI, uses Bayesian probability to interpret the general meaning of its machine-read words). Slang, sarcasm and idiomatic expressions obviously screw things up. On the bright side, the pool of data is growing exponentially from day to day, so researchers have no shortage of raw material to analyze.

The Trendrr crew thinks that as the science stands now, you can get a perhaps 80% accurate sentiment reading -- positive, negative or neutral -- on Twitter topics that have a critical mass of tweets.

All that said, a few notes about the latest Trendrr Chart of the Week, which, for the first time ever, is sentiment-analysis flavored:

• "Neutral" sentiment has to do with tweets that are generally simply informative (e.g., "Colin Firth is on Leno tonight") and don't express an obvious strong opinion one way or another.

• If you're an NBC executive, you obviously want to see positive sentiment about Leno exceeding negative sentiment. In that regard, Leno can't catch a break on Twitter -- lately his negative tweets consistently outpace his positive tweets, sometimes by as much as a two-to-one ratio. (By contrast, David Letterman -- even with his creepy office-sex scandal still in full swing -- has had positive tweets outnumber negative tweets three times during the same period.)

• It strikes me that in addition to deploying the intertwining sciences of statistics, probability and linguistics, what we really need to explore in the social-media space has to do with behavioral science. Like, are people with negative opinions about Leno just more likely to tweet them than fans? You might think so -- we all tend to presume that cyberspace is a cesspool of snark -- but as The Wall Street Journal recently reported in a piece titled "On the Internet, Everyone's a Critic But They're Not Very Critical," the average consumer product review online scores about 4.3 stars out of five. This according to Bazaarvoice, a company that "provides review software used by nearly 600 sites." The Journal adds that "Amazon Inc. says its average is similar." Go figure.

Anyway, sentiment analysis is a work in progress -- but obviously a potentially powerful tool for media people and marketers. Stay tuned.

~ ~ ~
Dumenco's Trendrr Chart of the Week is produced in collaboration with Wiredset, the New York digital agency behind Trendrr, a social- and digital-media tracking service. More background here. A basic Trendrr account is free; Trendrr Pro, which offers more robust tracking and reporting tools, comes in various paid flavors (get the details here).

Simon Dumenco is the "Media Guy" media columnist for Advertising Age. You can follow him on Twitter @simondumenco.