On the Importance of a Hierarchy for Learning Continuous Vector Representations of a Label Space
Jinseok Nam, Johannes Fürnkranz · arXiv (Cornell University) · 2014
In multi-label classification, many attempts have been made to capture patterns of labels or underlying structures having an impact on such patterns. One of challenging tasks in this research is how to exploit hierarchical structures over labels. We present a novel method to learn continuous vector representations of a label space given a hierarchy of labels. Our experimental results demonstrate that the proposed method is able to learn regularities among labels by exploiting a label hierarchy as well as label co-occurrences. It highlights the importance of the hierarchical information in order to obtain regularities which make it possible to perform analogical reasoning over a label space. We also show that construction of such regularities depends on a given hierarchy.