If you had a buddy who was all-knowing, infinitely patient, good at undlerstanding your questions, and always available through instant messaging, you’d probably turn to him rather than a traditional search engine for information.
And that’s what Kozoru, a startup based in Overland Park, KS, wants to train you to do – except that the “buddy” is a piece of software. Kozoru’s technology, which launches June 5, turns instant messaging (IM) into a social search engine. Type a question into AOL Instant Messenger (AIM), wait for Kozoru’s system to survey authoritative websites, and see an answer materialize.
What appears is not a list of ranked pages with the fact you need buried deep inside, though. Rather, Kozoru (a Japanese word that means “to gather,” as in a band of samurai gathering to do good and combat evil) delivers a full-sentence answer, just as your buddy with the encyclopedic mind might. For instance, to the question “What was McHale’s Navy?” Kozoru responds, “McHale’s Navy was an American television sitcom series. The series ran for 138 thirty-minute episodes (along with a one-hour pilot) from 1962 to 1966 on ABC,” and then links to a Wikipedia page for more information.
As hope and capital coalesce around Web 2.0, one of the areas of greatest frenzy is social networking: media and merger-and-acquisition darlings such as Facebook, Flickr, and MySpace attract millions of visitors per month, who are eager to interact, share, and create content. Another area is second-generation search, in which heavyweights such as Autonomy, IBM, and Inxight are moving beyond simple key-word searches made ubiquitous by Google. Using sophisticated pattern-recognition software and natural-language processing, these companies are attempting to identify the meaning behind the words.
Kozoru’s technology is what Web 2.0 mavens call a “mashup” of these two trends. The company started in July 2004 by attacking the problem of natural-language understanding. “Computers are terrible at language problems,” says John Flowers, Kozoru’s founder and CEO. By statistically analyzing how words are used in human language, Flowers says, Kozoru can deliver relevant search results based not on key words but on entire sentences and the meaning behind them.
“If you ask a question, we will answer that question in a way you expect to be answered,” says Flowers. Ask Jeeves, now simply Ask.com, once marketed itself in a similar way, but its answers were generated by humans building question-and-answer sets. (Ask Jeeves cofounder David Warthen is on Kozoru’s board.)
Many search companies and other firms lay claim to advances in natural-language processing. A quick test of Kozoru’s product shows that its service is like most others – hit or miss. But one clear distinction is that Kozoru places Web searching within the medium of IM, which Flowers says alters the expectations and the experience of users. Suddenly, they’re typing in questions, not just a few key words, and expecting full-sentence answers.
Other companies have used IM as a vehicle for more limited types of searches. Kayak, a travel search aggregator, has released KayakBot for AIM, which searches for flight and rental-car information. The venerable Wall Street Journal has long had an AIM bot, WSJ, that will retrieve stock quotes when users type in ticker symbols. Yahoo has filed patent applications in the area.