A Look into Sentiment Analysis for Brands
Posted by August 26th 09
Are you popular? Do people love you or your brand? Is there an angry mob awaiting your tweets with pitchforks? Let's face it, people's opinion about your brand matters a lot in this social media age. Reviews and ratings is a sought-after feature for users and studies have shown that it does have an influence on perceptions and sales. Nothing groundbreaking here.
Measuring online sentiments for a brand is a time consuming but invaluable process. Many startups have seen a business opportunity here in providing people and businesses with measurement and analytics tools that would do just that - drawing the portrait of a brand's online perceptions. The New York Times has an in depth article on the subject that is well worth your time.
Newssift is one of these startups. They've put together an engine that gathers online information about a brand, then scans them through their algorithm. As explained by the New York Times, "the simplest algorithms work by scanning keywords to categorize a statement as positive or negative, based on a simple binary analysis (“love” is good, “hate” is bad)." Wonder what sorts of results you would get? I ran one of my favourite brands within the engine to test it out. Here's what it had to say about Apple:
As you can see, sentiments towards Apple are overwhelmingly positive and neutral. This gets broken down into more specific trends that tells us what works for Apple (iPhone, computer security) and what doesn't (management, iPhone apps, voice apps.) This analysis can be taken to Twitter too. Tweetfeel and Twendz are two sites attempting to measure sentiments on the popular channel. Here's what a search for Apple resulted in:
Of course algorithms can't measure all the subtleties of the human language, nor can they measure something other than English (at least for now.) Indeed, "that approach fails to capture the subtleties that bring human language to life: irony, sarcasm, slang and other idiomatic expressions. Reliable sentiment analysis requires parsing many linguistic shades of gray (NYTimes)." Nonetheless, this is an encouraging start for sentiment analysis. I'm looking forward to have these features integrated to search engines in the near future! What are you waiting for Google?
posted in Thought Leadership
July 2010
