Similarity-based Multi-label Learning

Multi-label classification is an important learning problem with many applications. In this work, we propose a similarity-based approach for multi-label learning called SML. We also introduce a similarity-based approach for predicting the label set size. SML is amenable to streaming data and online learning, naturally able to handle changes in the problem domain, robust to training data with skewed class label sets, accurate with low variance, and lends itself to an efficient parallel implementation. The experimental results demonstrate the effectiveness of SML for multi-label classification where it is shown to compare favorably with a wide variety of existing algorithms across a range of evaluation criterion.