Nanocrowd
Take a staggeringly huge data set, and reduce it to its constituent parts.
Statistics and abundant computing power make this easy.
(See: Netflix Prize.)Push and pull and mix those pieces into one salient decision: “Here, this is what you want.”
This is hard. It’s much fuzzier logic, and it’s a decision that most procedural algorithms hang up on.
(See: Netflix Prize. Again.)Profit.
No one’s gotten here yet. Step 2 requires either: a very lucky guess, or flat-out mind-reading.
Nanocrowd’s allure is in what it lacks: there’s no step 2. It does what computers do best: tosses and turns numbers until something meaningful appears.
This is a step forward for recommendation engines: teasing apart the themes and characteristics that constitute a body of content is useful.
“I enjoyed that; I wonder what else there is like it?”
Each film is made of little bits of style and substance from many others. Plot devices, music, art direction, actors, et cetera.
None of the three word descriptions from Nanocrowd will describe the entire film, but that’s precisely the point: they each encapsulate a key part of it. Plucking out those clusters of thematic goodness for me to choose from is magical: “here, which part of this film are you in the mood for?”
