The website, which launched in November, mines social networking sites using a computer algorithm, and is able to tease illness trends out of a sea of social media-data.
Sickweather is the brain-child of Graham Dodge, a 36-year-old Windsor Mill resident, who pulled in a pair of his Dulaney High School buddies, Michael Belt and James Sajor, to launch the company.
Dodge spent years in Los Angeles during the first dot-com bubble, more than a decade ago, working in Internet media businesses. Living in Maryland for the last several years, Dodge works full-time as a marketing director for a Timonium accounting firm.
But for the past year, he and his partners have been nursing the startup dream with Sickweather, a site that aims to provide consumers with real-time information about illness trends in their communities across the country. The team is currently looking for early stage investors, but have already gotten media attention in the United States and England.
The site could be a boon to health care providers and pharmaceutical companies, who can use it to advertise their products and services to a highly targeted audience: sick people who want to feel better as quickly as possible.
Dodge recently spoke with the Baltimore Sun about Sickweather's origins, its competition and business model, and the prospects for adapting the site to forecast other trends using social media data.
How did you get the idea for Sickweather?
About a year and a half ago, I was sick with a stomach bug. My first inclination was to find out if anyone had the same thing. I went on to Facebook to see if my friends were talking about the same bug. Sure enough, I found a friend. Based on my background in web development and online marketing, and knowing how the [application programming interfaces] work on Twitter and Facebook, I knew I could do that search on a broader scale with keyword filtering, and be able to tell what symptoms were going around.
Is Sickweather your first startup project?
No, back in the first dot-com bubble (in the late 1990s), I had a startup called Notfilms.com. It was a portal for users to send us amateur videos. People would send us videos and VHS tapes and we would have a person encode the video and put it online. We would get a dozen or so videos a week and do our best to keep up with it. People didn't have broadband connections back then. This was before YouTube. There was no platform or venue for anyone who wanted to show a video of themselves skating down a sidewalk. We ended up not surviving the dot-com bubble bursting. Our funding dried out. And then a couple years later, YouTube emerged. The concept was there but the audience and the technology wasn't ready.
How does Sickweather fit in with the science of studying outbreaks and contagion in public health?
Two of our advisors are from Hopkins, one is a professor and the other is a Ph.D. student. They have received their own press for their own research in using Twitter to track illness. We quickly realized we had some synergy in what we were doing. Our approach is validated by their own research and the research of others who use natural language processing [to study online public health trends].
Several years ago, Google released their Flu Trends tool, which can show and predict flu outbreaks based on keyword searches people are using for flu symptoms. But, aside from that tool, Sickweather doesn't really seem to have much competition.
That's right, surprisingly. ... There was nobody mining the social graph [online social networking websites] for indications of illness the way we were thinking of doing, and nobody doing it in real time. ... Not only are we not limited to just tracking flu, we can also mine the social graph in real time and search any symptom or illness that people openly talk about on social media.
I noticed that depression was one of the top three sicknesses in Baltimore last week.
Yes. And after we [the Baltimore Ravens] beat the Steelers, depression was the no. 1 symptom in Pittsburgh.
Can you get into forecasting sicknesses the same way TV weather forecasters try to predict future weather patterns?
Yes, we can. The 7 million tweets we've collected, we've refined down to 1.3 million qualified reports of illness, with 90 percent accuracy. So 9 out of 10 times, our algorithm is correctly identifying a sickness. One of the things we're rolling out soon is time lapse maps, so you can see where diseases are growing. It's very much like weather forecasting. It's early days for illness forecasting. We're really the first company to offer sickness forecasting for consumers. We were able to forecast whooping cough in Algonquin, Ill., about two to four weeks before it was reported in the media. In that sense, we're already predictive.
Where are most of your users?