In multi-label learning, each instance can be related with one or more binary target variables. Multi-label learning problems are commonly found in many applications, e.g. in text classification where a news article is possible to be both on politics and finance. The main motivation of multi-label learning algorithms is the exploitation of label dependencies in order to improve prediction accuracy. In this paper, we present ongoing work on a method that uses the linked open data cloud to detect relationships between labels, enriches the set of labels with new concepts which are super classes of two or more labels, trains a model on the enhanced training set and finally, makes predictions on the enhanced test set in order to improve the prediction accuracy of the initial labels.