Remember when computer scientists promised back in the 1990s that we’d all have artificially intelligent “personal agents” running cyberspace errands for us, booking travel, filtering the news, and finding things we like? A new widget for personal computers called a “MatchKey” doesn’t quite qualify as intelligent, but it does take a new approach to the third problem—sifting through the Internet’s vast catalog of content and picking out items that match your preferences. It has the potential to create far more accurate music, movie, and product recommendations than the ones Web users receive today, and it has emerged from the unlikeliest of places: the company that owns the New England Patriots football team.
At the DemoFall conference in San Diego yesterday, Needham-based startup Matchmine unveiled what it calls a “media discovery platform” built around MatchKeys—mathematical representations of individual users’ preferences in areas such as music, movies, and other online media. Once Web surfers have signed up for their own MatchKeys, the keys will monitor their media-consuming habits and continuously update their preference scores. Whenever users visit Matchmine partner sites such as Fuzz, Filmcrave, and Peerflix, software at the sites will tap into their MatchKeys and present content fitting their profiles.
“The big idea is that we want to make online media personalization portable,” says Mike Troiano, CEO of Matchmine, which also announced Monday that it has collected $10 million in Series A venture funding from The Kraft Group, the Foxborough, MA, company that owns the Patriots as well as paper manufacturing and real estate development interests.
The Patriots were among the first NFL teams to develop an extensive website with multimedia offerings such as a streaming radio station. Troiano says Matchmine grew out of a conversation in 2005 between Kraft Group president Jonathan Kraft and Trent Adams, the group’s “chief innovator,” about technologies that would help people find the Patriots’ content online.
“Trent realized that this is a problem for every user and every publisher: If you’re a user, how do you find interesting stuff, and if you’re a publisher, how do you connect with people who will be interested in your stuff,” Troiano explains. “Jon and Trent thought that if we could crack that nut, it would be a hugely profitable business. And within a couple of weeks, Trent had this epiphany about how to create a statistical representation of users’ tastes. That’s the core that became the MatchKey.”
Today, websites such as Amazon generate product recommendations using collaborative filtering software, which works on the principle of shared taste: “You bought this Annie Lennox CD, and a lot of people who like Annie Lennox also like Melissa Etheridge, so you’ll probably like this Melissa Etheridge CD.” But the basic algorithms behind collaborative filtering are nearly a decade old, and have many limitations, including the fact that they reinforce popular choices while marginalizing unusual or lesser-known content.
The information in a MatchKey, by contrast, isn’t about specific products or other people’s preferences. It’s essentially a listing of a user’s scores on 200 axes across different media types, including movies, online video, music, and blogs. Each media type has 50 to 75 axes; for movies, for example, an individual might have a low score on an axis such as “likes science fiction” and a high score on axes such as “likes mobster movies,” “likes violent movies,” and “likes movies with lots of celebrities.”
The other half of the equation is Matchmine’s “universal content catalog,” which ranks the items in its partner companies’ databases along the same axes found in the MatchKeys. When a MatchKey owner visits a Matchmine partner site, the site grabs the ID for the user’s MatchKey and sends it to Matchmine’s media discovery server, which reaches into the universal content catalog and sends back content tailored to that key. The individual mentioned above, for example, might get a recommendation for the movie Goodfellas.
To give users a sense of connection to their MatchKeys, Matchmine has also developed a graphical representation of the key, a spiky crystal sphere that looks like it was swiped from Superman’s Fortress of Solitude. “We want people to connect with their key on an emotional level and give them a sense of ownership,” says Troiano. The lengths and configurations of the spikes on a MatchKey are unique to each user, and while their meaning isn’t exactly clear, the company is working on visualization tools that will make the keys more meaningful to the general user. Says Troiano, “Some of our scientists can already take one look at a MatchKey and say, ‘This is a big sci-fi fan.’ But you can imagine tools like a 3-D flyover of your music zone with rollovers showing what each crystal means.”
Such baubles might be a nice distraction. But if there’s real value in a portable preference engine like Matchmine’s system, it’s that it could help consumers move to the next stage in media personalization—unifying their preferences in one place rather than leaving them scattered across dozens of separate e-retailing sites, where the last few albums or movies a user bought might be all the retailer knows about them.
Just as important, MatchKeys are designed to get smarter over time. “As you consume content in any Matchmine-enabled application, your key will evolve based on the preferences you express implicitly and explicitly,” says Troiano. “What that means is that the time you invest in any one discovery experience will benefit you in all the others.”
At the moment, there aren’t that many opportunities for discovery experiences through Matchmine, which launched with only three partner sites—artist-centric music sharing site Fuzz, user-written movie review site Filmcrave, and DVD-trading site Peerflix. But Troiano says Matchmine will be adding more partners “at a feverish pace” over the next few months. “What we’re finding as we talk to these folks is that they know they need to invest in the development of some kind of attribute-based recommendation system, but they just don’t have the resources or manpower, whereas we have a large organization dedicated to doing nothing but that,” Troiano says. “More often than not, they are eager to have us come in and score their content catalog and begin to make recommendations, because they know they have to do it.”
