Jennifer Zaino
SemanticWeb.com Contributor

The day will come — and soon — when web sites without video content will seem as old fashioned as not having a web site at all. That’s the philosophy driving Delve Networks, which figures that publishers are going to need a way to help their users navigate that video content and monetize it.

It’s not only a future problem to solve. It’s very real right now to companies such as NBC, whose news division as 50 million hours of archived video news content it believes to be worth over $1 billion in value. Or to the Colbert Report, which just spent 3 months and used 16 people to tag its videos for the web, says Alex Castro, Delve founder and CEO.

As content expands, it becomes almost unmanageable to rely on manual means to keep up with it. The special sauce behind Delve’s solution to this problem is running both speech recognition technology topped by a semantic layer.

Semantics is the angle others have missed, contends Castro, who worked on speech recognition back at Bell Labs years ago, gaining both an appreciation of it and a recognition of its limitations. Later he did graduate work at Cornell with Gerard Salton, the father of modern search technology. At Delve, he has the opportunity to match these two interests together to improve video search, navigation and management.

Delve regularly crawls and categorizes hundreds of millions of web pages, and uses statistical natural language processing to uncover the relationships between words, looking at the proximity of certain words to one another and how often they occur together. Because the crawls are constantly updated, new relationships are often created — for example, a year ago you might not have found a relationship between the words “Elliot Spitzer” and “prostitution.” Obviously, that’s changed.

“We think we are unique – the speech engine world has been so focused on making that smarter than no one took a step back to say, here is a big web with lots of data, and we can use that to make speech work better,” says Castro.

As an example, its crawling may turn up that there’s a strong relationship between Muqtada al Sadr and Iraq, information that can be put to good use when a user wants to key in on that segment in a 5-minute news clip that also discusses an ice storm in the Midwest and the mortgage loan crisis. Because Muqtada al Sadr is an unusual English term, it might be missed by the speech recognition engine but caught by the semantic relationship that was uncovered between Muqtada and Iraq via its natural language processing technology. A “heat map” then shows up over the region where the Iraq information is presented on the video clip.

“You’ll see that this 45-second chunk of that video is what you are looking for even if you can’t find the actual word,” says Castro. “That’s how a semantic knowledge of how words are related creates the perception to the user that speech recognition is better than it actually is.”

Not only does the technology, which companies can subscribe to as a hosted service, get the user where he wants to be without taxing the publishers’ time and resources. It also recommends other pieces of related content (videos or text) to drive more consumption and possibly monetization opportunities for publishers. “Without the semantic capability, these things aren’t possible,” Castro says. Current customers include CNet for their BNet and ZDNet sites, bikini.com, and more recently the Kansas City Chiefs.

Delve believes the market for its solution will be driven by trends including the popularity of YouTube.

“It’s clear to me that YouTube won’t host all the world’s video content, but it has helped push it in the extreme. Companies will use YouTube to drive back traffic but they want their own content, their own monetization, and video on sites five years from now will be the cost of doing business,” Castro says. “They need the tools to make that possible and the tools that exist today treat video as an opaque object. If you look at the text web, we’ve seen so much intelligence with Google making it fashionable to care about text processing. Now that we are starting to see the first glimpses of the semantic web evolving, we looked at that and said the video side has been locked out of that. We want to elevate it to a first-class data type on the web.”

Following the announcement of its complete platform solution a little over a month ago, Delve is planning soon to announce an analytics solution to let content publishers track what people are watching, how much, and what they are searching for in that video. This fall it plans to come out with the capability to go beyond searching within a single video to search all video at a site that subscribes to the service, then hyperlink to the exact segment of the video that matches the query.

Semantic Tech & Business Conference Returns to San Francisco

Semantic Tech & Business Conference returns to San Francisco in June! Join us from June 3-7 for complete coverage of Big Data, Linked Data, Extreme Information Management, and Semantic Web. From breakthrough approaches to solving business problems to the big data implications of fast–evolving technologies, SemTechBiz provides you with an unparalleled interactive experience and delivers tangible business value. We're offering a special early rate when you register by February 17. Sign up now!