The Web Has Feelings Too: How Seattle Startups Are Cashing In on Sentiment Analysis
Laugh, and the world laughs with you. Weep, and you weep alone (as the old saying goes). Or maybe not. Maybe no one weeps alone anymore, what with Twitter, Facebook, and the like.
This week’s release of a “sentiment analysis” product from Seattle-based Appature made me think about what’s really going on at the intersection of social media and semantic understanding of the Web. Appature is an up-and-coming, already profitable startup that has made its name in helping companies in the healthcare industry better understand their customers, so as to deliver more effective marketing campaigns. It has made inroads with big customers, like Johnson & Johnson and Microsoft Health Solutions Group.
Appature’s latest offering “allows marketers to close the loop between the pulse on the broad Web and their non-Web business in an immediately actionable way,” says Kabir Shahani, Appature’s co-founder and CEO. “It’s the linkage between the broader Web and the other information you already have about a customer or prospect, with immediate action built on top of this, that is the key here.” In other words, it’s not just about market awareness, it’s about directly driving revenues—by monitoring and understanding positive and negative feelings about brands, products, and other industry topics, gleaned from blogs, Twitter, Facebook, other social sites, and Web documents.
It’s a fast-moving trend, and there are now enough Seattle companies working on this kind of technology to put the Northwest squarely on the “Web sentiment” map. (I haven’t talked to companies in Portland, OR, yet, though Jive Software comes to mind.) Here’s how it all works, in a nutshell. If you’re a marketer, you can buy software to help monitor and summarize what customers are saying about your brand on the Web. If you’re a big company, or an individual with a reputation to protect, you can even do things like change the visibility of search results that pop up on your name. And if you’re a regular old consumer, you can keep track of things like what percentage of people (and who) are saying positive or negative things about the iPhone, Microsoft, or Kanye West.
That kind of software is exactly what Seattle-based Evri rolled out in August. Evri’s calling card is that it scours the Web every minute of every day, trying to understand the semantic meaning of sentences on Web pages and documents. What’s more, its software makes connections between all the entities (names, places, products) it encounters on the Web. The broader goal is to give people a better way to browse and discover information.
“One thing we do which is really different is we actually say who or what is expressing the positive or negative sentiment, as opposed to simply remaining ambiguous,” said Deep Dhillon, Evri’s chief technology officer. “One of the reasons we can offer these more involved capabilities is due to the depth of our technology stack—because we know, for every sentence, who the subject, verb, object, etc. are.”
Which raises the question of just how far along the semantic technology really is. Philip Vaughn, co-founder of the Seattle-area startup Raveable, has some valuable perspective. Raveable is a travel website that summarizes the sentiment of hotel user reviews. As Vaughn explains, there are three different technical approaches involved in these various companies: making sense of opinions (sentiment analysis), classifying topics (what is a statement about), and mapping relationships (who else has similar sentiment). “From a business perspective, there is an opportunity to do [things] on a scale never before possible,” he says.
An earlier conversation with Vaughn convinced me that understanding the sentiment of a sentence where the computer doesn’t know the exact context or topic—hotel rooms, say, or doctors in hospitals—is still too difficult for today’s commercial software (indeed this can be hard for a person too). It’s a deep research problem in artificial intelligence and machine learning, and many groups are working on it. “Broad sentiment typically has lower accuracy than topic-based sentiment analysis,” Vaughn says.
But, for certain targeted business applications, sentiment analysis has clearly found a home—especially when packaged with other customer-focused products. Take Appature and the healthcare market. “From what we’ve seen of other offerings, they do a good job of giving you sentiment around social media based on keywords,” Shahani says. “But it is not integrated with the broader set of data a company has about their customer. In many cases that satisfies the need, but our customers are looking to leverage social media in a much more actionable and relevant way.”
On that thought, the granddaddy of local “sentimental” startups (my usage, not theirs) is probably Bellevue, WA-based Visible Technologies. Founded in 2003, Visible takes all of this a step further in another direction: reputation management for brands, products, and people. Back in June, the company released TruReputation, a software product that expands on what Visible is already known for—namely monitoring the Web for buzz about a particular brand or company, and helping marketers respond to it. The latest software finds positive and negative content about a brand online, say, and uses keyword placement, optimization, and linking techniques to make the positive entries pop up higher in search results (on Google, Bing, Yahoo, and AOL Search). How exactly it does this is a company secret, but Visible has certainly been getting a lot of interest.
So where is all of this headed? The business applications of sentiment analysis are clearly taking off in certain targeted marketplaces, and the Seattle area has become a leader in the space. Beyond that, the effort to make sense of greater and greater reams of data will continue, but semantic technology needs to improve before we’ll see it everywhere.