In this work we propose a novel attention-based neural network model for thetask of fine-grained entity type classification that unlike previously proposedmodels recursively composes representations of entity mention contexts. Ourmodel achieves state-of-the-art performance with 74.94% loose micro F1-score onthe well-established FIGER dataset, a relative improvement of 2.59%. We alsoinvestigate the behavior of the attention mechanism of our model and observethat it can learn contextual linguistic expressions that indicate thefine-grained category memberships of an entity.
展开▼
