Interesting concept. I suppose there are 2 elements (1) predicting good games and (2) watching replays of good games.
(1) Inputs to Predict Games:
a) Ladder Position relative to each other and Ladder position overall - pretty easy to get
b) Number of star players - to put into an algorithm either count number of all star appearances of playing rosters OR check player salary from say a site like hoopshype.com and see how many players earn over $15m per year in the game
c) Trending track record - check last 10 games for both teams
d) Rivalries - not sure how to build data on this.
e) Fav Teams - user input into teams they like (and a negative list of those they don't
(2) Trying to Predict if a game was good:
a) Social Media coverage - number of tweets, FB posts, links to youtube (this will be skewed by Kobe and LeBron fans/haters) but if good plays happen or good games they are shared on social
b) Margin at the end - I've seen good games have a team lead 95% of the game and lose or a team blow out one early and then come back to win. Rarely do I think a game has been good if the last quarter is pointless and all the scrubs play
c) Number of total points - a 65-66 (aka Memphis) game with little highlights isn't as exciting as a 132 - 120 game (aka Golden State).
In order to predict the game you might also want to check out seatgeek.com as they have an algorithm to check how much your ticket is worth based on the teams playing, their laser position, the weather, your seat location, etc.