Google Analytics gives web site owners good information about what’s clicking with visitors to their site, how those users got there, and more. But, attaining that insight can be somewhat laborious for those not well-versed in the tool and its interface, says Nick Cassimatis, associate professor in the Department of Cognitive Science at Rensselaer Polytechnic Institute. He’s the founder of natural language processing technology startup SkyPhrase. His take: Apply SkyPhrase to the task, and things get a lot easier.
The startup in February began private beta testing of its NLP interface to Google Analytics. “Google Analytics lets you ask things like how many people from California visited the site last month, or which of your pages were most visited on mobile devices,” says Cassimatis. “Our system lets you ask these questions in natural language and get answers to them” more seamlessly than using Google Analytics alone.
Previous to bringing its NLP help to Google Analytics, SkyPhrase had a public site that let users run natural language searches of their Gmail or Twitter accounts, as well as flights and music.
Its technology, for example, could understand Gmail and Twitter queries missed by Siri, Cassimatis says, such as “emails tagged receipts from Southwest the last three months” or “PDFs John sent with subject ‘resume’,” as well as “yesterday’s tweets about the debate” and “search twitter for pictures of NASA.” Flight queries SkyPhrase understands that Google does not included items like “Delta flights from San Francisco to Los Angeles that leave tomorrow and return on Friday,” and “next Delta flight out of Chicago to Boston.”
“If you could just tell your computer what you wanted to do, that would save time. And that is the goal in natural language understanding,” he says. Siri got the ball rolling in a big way around social search. But, he says, there are still drawbacks in most NLP apps, in that they tend to understand simple queries best, and “they are narrow and brittle and expensive to develop.” One of the largest software teams at Apple, he notes, is the one working on Siri.
How is SkyPhrase different than other NLP players? “It lets you ask more complicated queries, do more complicated things, understand more complex language with greater precision, and [with a forthcoming API release] it makes it easy for third parties to develop natural language interfaces for new domains,” he says. In experiments with the API, he says, people without programming skills were able to put together a natural language interface that exceeds Siri or Facebook search within a couple of weeks.
SkyPhrase was born out of the work done by Cassimatis and his research group at RPI. The team’s careful study of linguistics and original artificial intelligence (AI) research led to advances in its AI algorithms that made them more suited towards how it modeled language scientifically, he says. “We did a really good job of developing AI algorithms that were suited for language, whereas a lot of other approaches basically wouldn’t be able to make these mutual advances together in linguistics and AI.”
As far as the private beta test it’s running using SkyPhrase on Google Analytics, progress is good, he says. “It’s a very small sample of people but we find they use it more and do more complicated stuff,” he says. In the near term, he’d like to find other areas where natural language can help with otherwise complex tasks, such as analytics, as well as enable a developer platform. And longer term? “In several years we want to be this platform for all the world’s data and services, that they are being used using natural language. If you have data, a web site or service, you can surface it using natural language and using SkyPhrase,” he says. “We want to be the platform people develop for that, and that they access data through.”
- Digirati and Semantic Web Company Innovate Semantic Publishing
- Big Data & Semantic Search: How They're Affecting the Job Market
- Daedalus Calls for Development of High-Level Semantic APIs
- MOLTO: Improving Online Text Translation with Machine Learning