The Advances in Multi-label Classification

摘要

Traditional single-label classification in machine learning and pattern classification fields is concerned with learning from a set of examples that are associated with a single label from a label set. While in some application fields, such as text/audio/video classification and genome/protein function classification, the examples for learning are associated with a subset of a label set. The advances in the area of multi-label classification are summarized and organized into two classes according to their strategy. Meanwhile, the main characteristics of these methods are described. Specially, the ensemble methods for multi-label classification and methods for multi-label dataset with new characteristics are discussed. Moreover the future research directions are pointed out.