While Facebook is busy amassing questions and answers from its 500 some million members, a new service called Swingly, which is coming out of stealth on Tuesday, is thinking a little bigger. Like the entire Internet bigger.
As described to me last week in a call with Swingly's managing partner Andrew Hickl, Swingly is a machine-generated answer engine that contains somewhere around 100 billion to 150 billion question and answer pairs. "It's the first one of its kind," Hickl said. "Our gimmick here is that Swingly uses our system and the Web."
What that means is that users can search for answers to their questions not from existing user-answered questions as found on places like Facebook or Yahoo Answers, but from sites where the answers can be extracted. That includes news stories and blog posts, as well as the comments that come with them.
The technology that powers this comes from the Language Computer Corporation (LCC) in Dallas, Texas, where Hickl resides as CEO. For the past 15 years, LCC has provided the government and other clients with ways to sift through large volumes of data using its semantic language processing technology. As Hickl joked, "We've been the go-to company that the government turns to when they're not happy with the natural language solutions out there."
Hickl says the technology is so robust that it can recognize around 15,000 different types of things. "That includes names and phrases. Everything from video games to left-handed relief pitchers." In the case of Swingly that translates to pairs of questions and answers, which the technology is able to link together and organize inside an index. "We take the text, find out what's meaningful, then flip it around for a Q&A pair," Hickl said.
But machines aren't perfect, which is why Swingly takes a hybrid approach to improve its answers. If someone is searching for an answer, and they know more than the system, or discover a better answer from one of its links, there are ways to suggest changes to the results. Similar to Google's now-defunct SearchWiki project, users can "like" certain results over others. To take it one step further, there's also a way to add in your own answer. To do this, you click the "edit" button on any answer it's pulled up. You can then write in whatever you want, and it's linked up with your account and will show up on the result just like a comment, just like I've done below:
One important thing to note here is how Swingly orders its results. For a search engine like Google, this order is called "relevancy." In Swingly's case, Hickl described its ordering as "trust," though noted that the idea of how trustworthy answers can be is difficult depending on what and where the content is coming from. "Numbers are a tricky thing. You get conflicting results based on when they were published, and the authority of the source. One thing we're trying to roll out is this idea of trust. Do I trust it? And does the community trust it?" To that end, there's no idea of what answers or sources are the best. Instead, users see what the system believes to be variations of the original question and links with previews of what they'll find there.
Along with the idea of trust is how timely the results are. "We grab temporally relevant information," Hickl said. "If I want to know who Don Draper's secretary is, we know that." Hickl added that for a question like that, Swingly would be able to point to other incarnations of that same question, which would hopefully point users in the right direction.
What if Swingly doesn't have any answers though? There's an answer for that in the form of something Hickl and company call "compare," which lets users ask the same question to both Swingly and a dozen other sites. It then takes the two sites and sticks them side by side in the same browser window. Considering how often many of the competitive sites actually show up in Swingly's own search results this doesn't make it a big draw, but it can speed up a search if you're in a hurry:
Swingly is being launched in private beta, though CNET readers who want to give the service a spin ahead of when it's made available to the general public can sign up with the invite code "CNET" (you have to type this in without quotation marks, and the characters are case-sensitive). In the meantime, Hickl said the company is working on improving its results and the speed, which is definitely slower than most other search engines. "One of the things we're going to be questioning is how long people will be willing to wait for better information...We'll be down to about two to three seconds when the site launches."