抄録
Multi-label classification has become a challenging topic. It assigns a data object to multi classes by associating a set of labels. The similarity between data objects is decided by the distance between their label sets. This paper discusses the similarity types and compares the distance measures between label sets. The similarity types are discussed from two perspectives, content and density. It shows that there is not a distance measure which can overall perform better than any other measures, so that different distance measures should be employed according to the condition of similarity.
