The paper analyzes several multi-label reduction approaches, and show the performance measure related to each approach. The reviewers agree that some of the results are novel, solid and non-trivial, such as how precision@k is optimized by pick-all-labels reduction and recall@k is optimized by pick-one-label reduction. While the empirical results are somewhat limited, the reviewers agree that the theoretical contributions are sufficient to recommend acceptance. Given that the results in the paper are clean and can be easily summarized, spotlight presentation is recommended. The authors are encouraged to improve the clarity of some parts of the writing.