The semantic web is an innovation upon an innovation. As this blog regularly covers, innovations are being built upon its broad shoulders all the time. But what is the inspiration for the innovators who see the potential in using semantic technologies in new and creative ways? And how do they continue to drive that innovation as their services gain traction and their start-ups begin growing?

The SemanticWeb Blog thought this was a topic worth exploring. So this week, we are, having raised the question with a few minds behind some recent semantic web start-ups, and we hope to continue bringing such perspectives as part of our regular coverage going forward.

We’ll begin this week’s series with an interview with Shyam Kapur, president and CEO of TipTop Technologies, which has developed semantic social search engine TipTop Search. As we covered,TipTop is a Twitter-based real-time search engine to let people find information they need by extracting data matching their search terms through its ‘reads’ of tweets. It groups what it extracts according to positive and negative sentiments to help guide people in making decisions.

TipTop’s founder and CEO has a background in the search world, having held positions at Yahoo, AltaVista, and Infoseek, and at other start-ups including e-discovery vendor MetaLINCS. Kapur says that the common theme to his work has been extracting knowledge out of unstructured data. That fascination clearly was the breeding ground for what ultimately became TipTop. But an equal fascination that propelled his original vision, and continues to influence how he tackles the ongoing direction and development of the site, is something that served as his research topic for his PhD from Cornell: How children learn language.

Connecting the Dots“It’s magical to acquire something as complex as language,” he says. “We’re not taught it, we just learn it.” How does that connect with manifesting his vision for TipTop? To the best of his ability, he’s trying to mimic with TipTop the way children learn language. As he sees it, “TipTop is a child’s brain to begin with. I believe that a child’s brain is not a tabula rasa, but it has some innate understanding of what is possible in language. And a child learns by being exposed to what he hears.” He fashioned TipTop to learn the same way, he says. “Let the data flow through the engine, and it is a self-learning system; it learns about what it’s being exposed to. Sometimes it learns immediately, and over time, with some prior knowledge, its understanding of an area gets even richer.”

The approach, he thinks, lets him avoid the issues of “having to invest in building sophisticated taxonomies and knowledge bases in particular areas, and then after years of work you get some meaning out of the text in that topic area. The big difference between TipTop and other semantic solutions is that we are able to apply across the board to any topic area.”

The most recent topic additions were sentiment trends around President Barack Obama (with a razor thin margin between satisfied--49 on TipTop’s satisfaction scale as of March 5, and dissatisfied at 51), and starting last week, the Oscars, tracking nominee sentiment trends and award predictions based upon tweet sentiment analysis in social media. (Avatar was the frontrunner in the Tips Leaderboard for best picture, Jeff Bridges for best actor and Carey Mulligan for best actress as of the Saturday before the big show -- one out of three ain't bad).

Kapur and his service are influenced by the idea that the algorithms in a child’s brain must be “cognitively plausible, and not require a supercomputer. So the way TipTop works is that instead of building huge super-computing models of all data, we tried to get the engine to read language like humans read language.” The advantage of taking this approach, he says, is it lets the service show evidence to support the claim the engine makes. “With other approaches you build vectors, compute classifications, you may even say this information is 80 percent about sports. But if you ask why it’s about sports, there is no answer. By doing it as I do, in a cognitively plausible way, when any quantitative score is reported, the supporting evidence is also in human-readable and understandable form,” which TipTop dubs the Tips and Pits ratio.

Kapur shares one thing in common with others who’ve launched their semantic web dreams – their efforts happened, in large part, because of their own frustration with the status quo, in his case, of finding information on the web needed to help make decisions. “I noticed the pain I felt as a regular user, then I looked at the technologies and said if you really want to do good job on making machines help make better, faster and more confident decisions, we have to make the machine do more with data than what search does. That motivated me to build the engine to read the content and understand it.

“When I say read and understand, to be clear, the difference is that others have AI inclinations to make the machine intelligent –that’s not my interest and that’s not practical. My way to solve this specific problem is what I can extract out of data that people could then leverage to make better decisions.”