The Debate About Automated Sentiment Analysis

In an earlier blog entry I described changing attitudes toward automated sentiment analysis, and I provided a recap of the recent Sentiment Analysis Symposium. There's more to say on the evolution of attitudes toward automated analysis. I will take up the topic where last Friday's blog left off.The Debate

So who are sentiment analysis's blogosphere critics, whom I referred to but didn't name in last week's article, the folks who are fighting a rear-guard action against automated technologies? There's Mikko Kotila, whose blog prompted me to ask Is Sentiment Analysis an 80% Solution? and who followed up with a March 31 response, Why Sentiment Analysis Sucks for Social Media Monitoring. There's Jason Falls, who lays out his reasons Why You Shouldn't Trust Automated Sentiment Scoring in his blog. Also some of the folks who posted comments to Falls' blog, although others invited Falls (and his readers of course) to learn a bit more about their companies' automated solutions.

Debate is healthy even if sometimes contentious -- let's illustrate it with a sketch, above, by Honoré Daumier -- and reiterate that regarding sentiment analysis, clearly the question is not If, it is How: automated or manual.

Actually, as just about any automation proponent would tell you, the better answer is Both, a hybrid, automated-human system. Let the machines tackle what they're good at: speed, reach, and volume, and also knowledge discovery and predictive modeling.

The widespread adoption of automated(-hybrid) methods is inevitable given automation's ability to vastly expand the scope, speed, and insightfulness of sentiment efforts. I see additional Conversions of the Critics, à la K.D. Paine's, coming soon. When "seeing the light," per Constantine's Conversion by Peter Paul Rubens, is inevitable, the question isn't even How, it's When.

Other observers, unlike the ones I cited above, are not entrenched defenders of human-only sentiment analysis. Check out Priyank Mohan's Sentiment Analysis: Can you get it right by just automating it? and Jennifer Zaino's Customer Sentiment in Social Media: Massive Scale is Scary, And Just One Link in Larger Chain of Data, which presents a nice review of recent market developments including Attensity's planned takeover of social/media monitoring platform provider Biz360. See also How Companies Can Use Sentiment Analysis to Improve Their Business, by Maria Ogneva, Director of Social Media at Biz360, posted at Mashable a couple of weeks ago, before the Attensity deal was announced.

(By the way, expect more deals like that one, soon. I'd put even money on Clarabridge's announcing their own social-media acquisition in the next week or two.)

There are some very helpful discussions out there, for instance in the replies to an inquiry posted by Greg Padiasek, a software developer at RIM, to the LinkedIn Text Analytics group. (You'll need to log in to LinkedIn, and you may need to join the group, to read the thread.)

Moving Forward

As for me, I presented a talk, Search for Sentiment, just last week (April 27) at the 2010 Search Engine Meeting, which was organized by analysts Stephen Arnold and Sue Feldman and included a number of really instructive talks. (Disclosure: Steve paid me a dollar for asking a question after his talk.) Looking ahead, I expect there'll be a strong dose of sentiment content in this month's Text Analytics Summit, May 25-26 in Boston. And I am working to replicate the sentiment symposium in San Francisco in the fall -- drop me a note if you'd be interested in participating -- this time, I hope, with additional representation of the listening-platform crowd. Why?

The Conversion

Listening-platform sentiment capabilities are basic for the most part, and even where they're more sophisticated -- providers such as Biz360 (morphing into Attensity360), Radian6, SAS, Scout Labs, and Sysomos -- they too often operate in an online-media silo that excludes data (and insights) drawn from enterprise transactional and operational systems and from qualitative (and quantitative) enterprise feedback sources such as surveys and contact-center notes. Perhaps it's the narrow scope and focus of social-media analytics -- practitioners' seeming limited knowledge about BI, text mining, and predictive analytics and about enterprise analytics applications -- that's to blame for outdated misperceptions about the accuracy and applicability of automated sentiment technologies. As they understand how good the technology is becoming -- correctly and appropriately applied -- and as they understand the use-cases beyond social-media measurement, I expect they'll join the ranks of automation believers.Attitudes toward automated sentiment analysis are changing as use cases and user stories are getting airplay Yet there are still techno-skeptic hold-outs. Debate helps, and a variety of voices have weighed in...