The opening keynotes at this week’s Semantic Technology and Business Conference saw two industry giants pump up the volume about how, and why, to apply semantic technology in the enterprise.
At Viacom, the largest pure-play media company in the world, the sheer number of perspectives across an exhaustive portfolio that includes more than 160 networks and 500 digital media properties globally, as well as entertainment behemoth Paramount Pictures Corp., was a factor in giving semantic tech a start. Its pain point, chief architect Matthew Degel told attendees, involved dealing with issues like the creative variations that come with the territory – U.S. vs. international versions of digital assets, or the MPEG-2 take on a clip for broadcast in this country vs. H.264/MPEG-4 formats for streaming the same clip online. “How do you track all this and say that I have 23 files, they are all sort of different but they’re talking about the same thing,” Degel said. “We thought semantics could help address that.”
Multi-platform being the rule of the day, the company faced the challenge of making its material reuseable, findable, searchable and purposeable, Degel said. As it takes steps to its goal of providing a corporate-focused, general purpose application of the technology, Degel explained that the view he takes on semantic technology is to think of it as “helping you deal with a certain amount of uncertainty and chaos.”
Degel took the audience through a tour of how to sell semantics up and down the corporate chain, from the C-level execs who are willing to okay big projects if they can understand a technology’s conceptual advantages to transform a business, or generate new markets or revenue streams, through to tech staff levels and developers. In that latter group, he said, you should “find those people who are superstars, deeply immerse them in this to create library, frameworks and extraction layers so other developers can just call into a service.”
And, as projects move along, be sure to measure your progress to build traction. Do it by the effectiveness of your ontologies, the utility of utilities you have created and the world views being changed by semantic stores. As an example, at Viacom Degel and his team worked on a semantics-driven app for its facilities group to help them track people’s office moves, using a semantic approach to bring together data from various silos like HR, phone, and facilities to know when a user logs into a new VoIP phone, who that user is and where that log-in places them. “It’s a simple problem to solve but it seems like magic to the people we deliver it to,” he said.
And, for managing media in the environment, as it relates to the joys of things like metadata tracking and different internal communities’ ideas about what media is (one man’s episode, after all, is another man’s installment), Viacom has been very successful in creating a media model “that for the first time integrates all perspectives,” Degel said, through to those of distributors like Netflix and iTunes. “We’re now looking at how to branch that further afield,” Degel said. “It was a tremendous victory for us.”
Yup, pieces of the puzzle are still missing to go Rambo-enterprise, like standards that aren’t followed by all, tools that aren’t entirely out-of-the-box-ready, security that isn’t as locked down as it needs to be, and debugging and troubleshooting that, he said, is “still far below where they need to be to take off in the corporate environment.” But for those who have a sense of the problems they need to solve, that shouldn’t be a deterrent to adopting semantic tech.
Walmart Weighs In
Degel followed Abhishek Gattani, senior director at Walmart in the WalmartLabs, who’s leading efforts to deploy semantic technology in his work at the division, which was created as an innovation engine for the mega-retailer. The Semantic Web spoke with Gattani in advance of his keynote (see story here), and Monday’s session brought more details into focus about the application of semantic technology at a company that counts 200 million visitors walking into its stores each week and 43 million unique visitors to Walmart.com per month.
Gattani expanded on the point he made in the previous article about the value of looking outside the business for data, in discussing, for example, the retailer’s effort to get people right to red shirts if that’s what they’re querying for, including appropriate visual representations. To that end, it developed semantic algorithms for color detection to, in red’s case, understand that it is a color, to rank shirts that come in red first, to understand what colors are close to red and show them if red is not available, and pick the right accompanying image.
“We were feeling really smart,” he said – until the realization that the same approach would be problematic if, instead of searching for a red shirt, users were searching for Red Dragon, for example. They’re not likely to specify they mean the movie, and, he said, you don’t want to show them dragons that are red. That’s where external data enters the picture: “Could we understand these things by looking at the web? [Things like] Red Dragon and Green Lantern are concepts in Wikipedia, as movies and entities, whereas red shirts are not,” he said. “So if a query is a topic in Wikipedia we don’t apply the algorithm. The big insight was going to an external data source to help.”
External data comes to the rescue again to help with issues like directing site visitors to similar products if they’re searching for something that’s not in stock, or in helping point them to the exact right product they want when there may a variety of products that all employ the same name in some way. For instance, if the Walmart search engine understands that a Kindle (which Walmart doesn’t sell) is a tablet and a device used for reading, it could leverage that knowledge to point visitors to products Walmart does sell, like the Nook. Or, if it can understand that it’s currently the Halloween season, a search for Batman should pull up costumes of the Caped Crusader, not DVDs of the Batman movies. “The user doesn’t expect to have to work around the experience,” he said.
Twitter helps there, with a caveat. Searching for Kindle Fire or Kindle in tweets, for instance, can point to words that co-occur with it, like iPad or Nexus 7, “so if we carry those products we can derive more meaning from those,” Gattani said. But have a care to filter for tweet spam before enjoying the fruits of named entity recognition, classification and associative rule mining around Twitter data. “Once that is done,” he said, “you can return products that are more relevant, like the Nook.” He also discussed how important it is to leverage event detectors that combine external web crawl and Twitter data to help understand things like sudden decreases in internal web site traffic for a particular item. The more you know, the better to form a complete picture. “Often,” he said, “incomplete information is more dangerous than no information.”
That’s just the start of semantics at Walmart. When it comes to search stream type ahead concepts, work is underway on rendering a new feature with semantic concepts in its quiver, he disclosed. It’s working to categorize queries while people are typing in order to direct them to the department that’s really most interesting to them – to electronics, for instance, if they’re searching for an iPad rather than to books, which may include publications written about the iPad.
“If I see iPad in books as the first recognition that’s bad, because it’s most likely not what the user wanted,” said Gattani. “We employ lot of machine learning to build models for queries, if they are ambiguous, about which are the different intents.”