[adventured]

Look good, but it needs to assess sentiment rather than just buzz (and it needs to grade buzz a lot better).

Gangster Squad for example has a 80 buzz score, but it got between terrible and mediocre reviews. Django has an 85 buzz score, but got almost universally great reviews. Django will make roughly 4 to 5 times as much money as Gangster Squad, and was just about blanket everywhere in the media 24/7, so it's very unlikely Ganster has anywhere close to as much actual buzz as Django.

[F_J_H]

You're right, but your comment made me think of something Andy Warhol once said:

"Don't pay any attention to what they write about you, just measure it in inches"

[visarga]

I'd train a linear regression on the text features to approximate the box office returns. That would be a way to build a rating system - the value of the function is the rating itself.

[mrgordon]

Exactly. I'd love to see this with sentiment analysis (ideally done by humans instead of support vector machines)

The data from people using Senti to rank movie sentiment has been interesting: https://senti.crowdflower.com