The only website I think is worth watching for polling analysis is Nate Silver's fivethirtyeight.com. The folks there are just about the only ones more interested in being accurate than in spinning the polls in ways that aid their personal causes.
They've been trying to figure out a way to forecast the Academy Awards and they've failed…so now they're asking readers to come up with ways to approach this puzzle. My suggestion? They're wasting their time. Not enough data is available. And not enough data will ever be available.
The folks who administer these awards want it that way. So do the folks who stand to receive one. They all like the idea that people think the voting represents a genuine consensus of Hollywood. Knowing more about who votes and how can only diminish that impression. So that's why the awards will remain largely unpredictable except by hunches and hearing "the buzz."