Jointly Extracting Relations with Class Ties via Effective Deep Ranking. (arXiv:1612.07602v1 [cs.AI])

In distant supervised relation extraction, the connection between relations of one entity tuple, which we call class ties, is common. Exploiting this connection may be promising for relation extraction. However, this property is seldom considered by previous work. In this work, to leverage class ties, we propose to make joint relation extraction with a unified model that integrates convolutional neural network with a general pairwise ranking framework, in which two novel ranking loss functions are introduced. Besides, an effective method is proposed to relieve the impact of relation NR (not relation) for model training and test. Experimental results on a widely used dataset show that: (1) Our model is much more superior than the baselines, achieving state-of-the-art performance; (2) Leveraging class ties, joint extraction is indeed better than separated extraction; (3) Relieving the impact of NR will significantly boost our model performance; (4) Our model can primely deal with wrong labeling problem.

from cs.AI updates on arXiv.org http://ift.tt/2i7dI2c
via IFTTT