1. Greenbook 2
  2. Greenbook-Mobile-6.29.16-

A Blizzard Of Insights At The Sentiment Symposium

Last week's Sentiment Analysis Symposium packed an astonishing amount of high quality content from top flight presenters into a single day.

 

By Bill Weylock

Last week’s Sentiment Analysis Symposium packed an astonishing amount of high quality content from top flight presenters into a single day. Seth Grimes from AltaPlana deserves a combat medal for resourcefulness and dedication for replacing overnight four out of the six speakers who were blown off the roster by Sandy. In the end there were 24 presentations (including a number of short “lightning talks”). It’s impossible to cover all, but I urge you to reach out to Seth and also to attend his next outing. The food was great, too, by the way. The man can stage a conference.

Although the focus was sentiment analysis process and applications, the theme was a bit elusive since there was such a profusion of perspectives. In the aggregate it accurately portrayed the field: increasingly crowded, with exciting applications, intellectually engaging, and filled with ferment and enormous potential. Topics new to me were deriving sentiment classifiers from speech data (call centers) and nascent steps toward classifying untagged images by emotional content.

Leading off the impressive roster, Kate Niederhoffer of Knowable Research suggests that the best route to assuaging broad client concerns that sentiment is an unreliable predictor of behavior is not a frenzied rush toward refining analytics to deliver “true sentiment.” While that may have intellectual merit, the larger point is that sentiment is already a very useful KPI when properly understood as only one of a family of factors affecting and predicting behavior. The focus is more usefully directed toward a better understanding of how sentiment relates to other metrics. Most basically, whether it is a leading or trailing indicator, and what phasing is appropriate between sentiment and other measures. Her prescriptions include immersion in all data available, utilizing entity extraction to add to the story, and appreciate sentiment as valuable but not a shining beacon answering all issues.

Apologies to the many excellent presenters I just can’t cover in the space available. Here are others that stood out for me:

  • Carol Haney (Toluna) shared successful research methodology illuminating category behavior by requesting access to respondents Facebook pages in the course of a survey (much as iModerate offers interpolated IDIs).
  • Aleksander Sobczyk showed us how Thompson Reuters has approached parsing sentiment past positives and negatives to derive anger and gloom. While still a work in progress, they have been able to correlate gloom with stock market lows (no surprise there) and see fear as a leading indicator of market declines. The ultimate value of being able to scale euphoria to predict market peaks (and optimal sell time) is immense. Loved his mini-history of the company back to its roots as a bunch of carrier pigeons flying market movers from Germany to England.
  • Catherine Van Zuylen of Attensity showed how sentiment analysis can not only illuminate the past and present, but can also affect future outcomes if reacted to quickly or in real time. What, for instance, might have changed in the debate if Biden had been able to see in real time that his grins were generating negatives? Social media and sentiment analysis can enable organization to take initiative or react on a dime and change their future or their competitors’ futures.
  • We learned that Bill Tuohig uses sentiment analysis at J.D. Power to flesh out and make NPS more actionable.
  • Mike Moran gave two presentations, one in a lightning round on Converseon’s use of machine learning to provide context and clarity not achievable through word spotting and Boolean methods. The other covered Google’s heroic efforts to stay one or more steps ahead of spammers and spoofers trying to game their site rankings.
  • A particular favorite was a short presentation from Erin Olivo on how SmogFarm is classifying facial expressions and physical typologies into emotional categories that include envy, disgust, sadness, and anger. They were able to predict a Gallup poll result one day in advance. I want to know more.
  • Vaidyanatha Siva of Infosys delivered a fascinating presentation on the possibility of deriving a “720 degree” perspective on the consumer by gaining access to her Facebook pages and mapping a “consumer genome” that would include influencers, likes, dislikes, activities, behavior patterns, likes and dislikes of family members, among others. This would enable targeting the appropriate message at the appropriate moment, with the appropriate tonality. Golly.
  • Cisco is integrating sentiment analysis with big data from their own pools and from over 300 media sources, including financial and industry analysts, according to Elizabeth Rector. They are close to what they consider a holistic view of their categories and positions within them.
  • Chris Jones, in one of the last two presentations, showed how Zynga is using sentiment measured from social exchanges within their games and from external gamer hives to modulate game levels and raise Zynga’s own game in the marketplace.

It was a long day, and one of the best I’ve spent. Don’t miss the next one.

Share
You can leave a response, or trackback from your own site.

Leave a Reply

*

%d bloggers like this: