The paper proposes a number of interesting ideas, and includes really strong empirical results (and user study). However, the reviewers agreed that it could benefit considerably with more precise characterization and theoretical analysis. Please incorporate the feedback from the reviewers in the final version.