Hierarchy-Aware Entity Embeddings for Classification in Large Taxonomies

Authors

Smirnoff I. ,  Chistova E.

Annotation

Hierarchical classification is essential for organizing entities within deep taxonomies. However, deep hierarchical classification methods face significant challenges, including managing class imbalances in large, skewed taxonomies and limited controllability. This work introduces a novel approach to entity classification through hierarchy-aware representation learning, effectively addressing these limitations. Our method facilitates the efficient mapping of entity names into expansive taxonomies, ensuring reliable classification across both prevalent and underrepresented categories in real-world applications. Experiments in both healthcare and e-commerce settings reveal that our hierarchy-aware embeddings improve entity classification performance and offer greater controllability compared to previous methods.

External links

DOI: 10.1007/978-3-031-93612-8_5

Download the book (PDF) from Springer Nature: https://link.springer.com/content/pdf/10.1007/978-3-031-93612-8.pdf

ResearchGate: https://www.researchgate.net/publication/396480341_Hierarchy-Aware_Entity_Embeddings_for_Classification_in_Large_Taxonomies