From: kym@ukato.freeshell.org   
      
   Jorekunrin@gmail.com wrote:   
   > Hi all,   
   > I'm developing a location-based search engine for a few New York   
   > neighborhoods and was hoping that I could get some advice on natural   
   > language processing. Specifically, I'd like to figure out how to   
   > answers questions like these:   
      
   You, and a couple other guys. :)   
      
   > 1) "Where can I find the best pizza on the Upper West Side?"   
   > For questions like this one, I would like the engine to respond by   
   > searching for Upper West Side restaurants that users have rated highly   
   > for their pizza.   
   > 2) "Where do I go to have fun this weekend?"   
   > With questions like these, I would like the engine to use more   
   > questions to figure out what the user wants to do to prepare (e.g. "Do   
   > you want movie recommendations?" "Do you want restaurant   
   > recommendations?"). Once the user answers a few of these questions, the   
   > engine could give him specific recommendations. ("Yes? Well, here's   
   > where this movie is playing in your neighborhood").   
   > In both cases, the engine has to identify natural language patterns for   
   > questions. It should also be able to process more technical search   
   > query methods.   
   [...]   
      
      
   You have at least 1 immediate choice. Do you really want to solve the   
   problem, or just appear to solve the problem?   
      
   Sure, I'm know for liking to play with my food, but this is actually a   
   nontrivial question.   
      
   You can build up semantic networks for the domain in qustion, and build   
   all sorts of clever and bleeding edge search algorithms for it, but   
   are you willing to tinker around to come up with something neat and   
   not-so-powerful, or something like a hacker's nightmare and ready for   
   the marketplace by the end of the year?   
      
   Comp.ai -- going by the moderator, at least -- is populated by genteel,   
   well-mannered folk who get a buzz out of tinker with novel things that may   
   prove useful in a century or so. Or maybe never, if you become old   
   and cynical and just resort to induction.   
      
   OTOH, hacking up something that's based on simple analysis (e.g. Baysean   
   stuff) of a corpus of relevant Q&A is not very interesting,   
   but can fool (I -- of course -- mean this in the nicest possible and   
   non-Enron sense) the customers and anyone's boss that the system is doing   
   something for their 2c.   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)