Coffeehouse Thread

2 posts

Forum Read Only

This forum has been made read only by the site admins. No new threads or comments can be added.

Search Engine Contexts

Back to Forum: Coffeehouse
  • User profile image
    jsrfc58

    I have a generic search engine question. This was triggered by reading the article "Gates vs. Google" in this thread, but I thought it was a different enough topic to warrant a separate thread.

    The article discussed how Microsoft was scrambling with its own version of search after Google grew in size.  I think there is a larger issue here, though.  When are search engines going to make the next "great leap forward"?  I've heard about the semantic web, however, there are other approaches I am thinking about regarding the problem of context.

    I would think this issue could be approached at least a couple of ways...a) ask the user for context, b) suggest context for the user.  For example, if the user types in something like the word "chip" the engine could suggest concepts from "potato chips" to "computer chips" to "buffalo chips" or whatever is likely to be the most relevant based on past statistics (maybe even cough up a picture or two). Basically, force the lazy user to come up with a context--some way, somehow, but still allow them to turn the context "questions" off.

    How much is "context" taken into consideration as it is? I find it interesting how grammar checkers can be built for Word, yet there doesn't seem to be a great deal happening in terms of search engines (not counting ads), except "behind the scenes". Would it work to have a search engine ask you questions after you type in a specific phrase (let's say only words that would return an ridiculous amount of matches such as anything over a few thousand)?  I don't mean Google's spelling type "questions" if you make a typo on your search term.

    The questions would have the sole purpose of nailing down a "context" from the user.  For instance, in the "chip" example, instead of returning 1,000,000 hits, return with categorical questions (food chip, tech chips, etc), and repeat the process until the user gets tired of it or they have gone through a few categories in order to narrow the list of hits down. 

    On the back end, of course, you would have to work on things, too, but I would think for some words in the English language (for instance) you could come up with an interconnected web of "words" based on contexts sitting on a server setup that could be queried as needed.  Each relationship (word to word) would have to be weighted so that if you typed "potato chip" it would return Pringles instead of a link-heavy blog entry that starts "last night I tried those Olestra potato chips and man..."  I don't know if a "relational database" type setup as they exist today would be enough.

  • User profile image
    Shaded

    I'm not a programmer (as much as I would like to be and have tried) but I think I can take a stab at this one.

    Language is the context.  Ambiguous words do somewhat decrease the relevance and value of search results but the power of search engines is that language can be quantified measured and sorted en masse automatically.

    Ambiguous context, however, cannot.  If you make context based rules they aren't always true, just like they aren't always true in Word.

    Word only fixes the most common context errors, with a prompt - requiring user input. A search engine needs to automatically crawl content, and measure and sort without user input.  So for every user selected classification of context, there would need to be that many x the number of search results decisions necessary to be made by a human for a search to be generated.  Why can't the computer figure it out?  Well same reason it can't figure it out now, its not measured in the language - you need to measure something outside the language.

    ... that almost makes sense, yeah?

Conversation locked

This conversation has been locked by the site admins. No new comments can be made.