Wireless Web Wielding a Wireless Bot
Page 1 of 2
BOT 2001 banged its way into San Francisco yesterday, promising a bounty of bot-based babble on everything from SearchBots to ChatterBots and KnowledgeBots.
Part of the conference, which was hosted by internet.com, centered around wireless bots, an area many pundits believe represents the make-or-break technology behind a working wireless Web.
Dr. Michael Pazzani, CEO of bot developer AdaptiveInfo.com, is one such pundit. Currently on leave from a professorship at UC Ithrvine, Pazzani doesn't mince words when it comes to discussing the importance of bot technology in today's increasingly wireless world. In his view, it's abot darn time people take notice.
"Everybody dreams of a wireless Web that is fast and easy to navigate," he says. "The current reality is slow connections on devices with tiny little screens. Consequently, the only way to make a wireless Web really work is by creating bots that can learn individual tastes and then go out and automatically create a personalized environment that works off those tastes."
That doesn't mean, he stresses, turning to strategies that call for the selective dissemination of information. Anyone, he reminds, who has checked off a plethora of personal questions while registering for a Web-based service would shutter at the thought of attempting such a feat on their mobile phone.
"Selective dissemination is also not appropriate for the wide variety of topics the Web can offer," adds Pazzaani. "They are very coarse-grained; they filter instead of prioritize; and only about two-to-five percent of users fill out those check boxes anyway."
The real answer for wireless bots, says Pazzani, is adaptive personalization, something he considers the core offering behind his company's Adaptive News Server (ANS), a bot service that delivers news, classifieds, entertainment information and advertisements to wireless devices.
"VoiceXML and wireless constraints are very similar in that both prevent you from being able to scan quickly," says Pazzani. "But by analyzing what users listen to or read in the past we can employ the ANS to push similar stories to the top of the queue. Our service also recognizes users who have used the system before, so it doesn't give instructions every time, which is something anyone who has ever called an automated airline reservation number would appreciate."
Training the bot in Pazzani's world simply requires using it. For example, anyone logging on to an AdaptiveInfo-powered news service via their handheld (the LATimes.com is one) might surf in and ask for sports information, after which a list of the top sports stories would fill the screen. If the user reads a story on football and then goes back and requests sports again, two more football stories might show up, with an added golf story to support diversity.
"We call this short term personalization," explains Pazzani. "Basically it's just giving the reader more stuff that is like everything they've read recently and less stuff that they don't appear interested in. That 'less interested' stuff is still there, of course, but it's just filtered down to the next few screens on your handheld or whatever device you might be using."
Long term personalization, which the ANS bots also employ, comes from recognizing similar words in stories users tend to read. So if users happen to read lots of stories that mention "dotcom" and "layoffs," the bot would recommend further stories that also mention those keywords (and hopefully recommend a therapist and pieces on career planning as well).
"We know personalization works," stresses Pazzani. "During a month of our own testing, users who were given personalized news stories on their mobile devices read 145 stories, while users who didn't receive personalization read only 100."
Pazzani firmly belie