4 ms·
Current search would be pretty embarrassing to future generation as much as it is awesome to current generation. Right now you can find a page only if that page
by sytelus 13y ago
Current search would be pretty embarrassing to future generation as much as it is awesome to current generation. Right now you can find a page only if that page contains all keywords in it. In some cases, there are band-aids to include popular synonyms and correlated keywords or remove some keywords or suggest alternative keywords (spell check). If you think of it, this is very primitive, pretty much the same stage as baby tryieng to sit up at 6 months.
Let's say you read the book Black List by Brad Thor and you had an awesome memory to remember everything. Your friend won't ask you about this novel by keywords. They will ask you questions like "What is the best way to steal a car?" or "How would you get rid of strangers chasing you?" or "what breed of dogs are very vicious when you signal but very friendly otherwise?". And you would be able to answer those questions, not only by quoting related text but also summarizing and enriching the information. This what the ideal search would look like... pretty much how you can converse with computer on Star Trek. That's why I said current search is just "keywords in a textbox", still a very far thing from making science fiction a reality.
- derefr 13y agoOh, I see where I was confused; you're not actually talking about "search" here. "Search" is the ML term for, basically, "finding something you previously stored, by using parts of it." Google is a search company. Search problems ultimately boil down to the simulation of an ideal http://en.wikipedia.org/wiki/Perfect_hash_function http://en.wikipedia.org/wiki/Perfect_hash_function generated from all the relevant data ever input. Search is an objective problem, best solved with lots and lots of Big Metal. This, on the other hand, is recommendation--finding something you don't know you want, by clustering things together and modelling your desires and personality. Amazon and Netflix are recommendation companies. Recommendation problems ultimately "boil down to" a full Bayesian network modelling your own brain, so they can have the right context to answer you. Recommendation is a subjective problem, best solved, probably, with a neural implant or something which can literally "read your mind" to know what you would like the most even if you don't currently "want" anything in particular.