ICNLSP 203: Representation Learning for Hierarchical Classification of Entity Titles

Title of the presentation: Representation Learning for Hierarchical Classification of Entity Titles.
By: Elena Chistova, FRC CSC RAS, Moscow, Russia.
6th International Conference on Natural Language and Speech Processing.
icnlsp.org/202...
Abstract:
We present a method for effective title encoding for hierarchical classification in a large taxonomy. The method enables taxonomy-aware encoding in pre-trained text encoders, such as fastText and BERT, which are additionally fine-tuned for the hierarchical classification. The embeddings produced using our method perform well when applied to nearest neighbor classification. They allow for controllable and sufficient hierarchical classification based solely on the title.