Distantly Supervised Relation Extraction with Sentence Reconstruction and Knowledge Base Priors

We propose a multi-task, probabilistic approach to facilitate distantly\nsupervised relation extraction by bringing closer the representations of\nsentences that contain the same Knowledge Base pairs. To achieve this, we bias\nthe latent space of sentences via a Variational Autoencoder (VAE) that is\ntrained jointly with a relation classifier. The latent code guides the pair\nrepresentations and influences sentence reconstruction. Experimental results on\ntwo datasets created via distant supervision indicate that multi-task learning\nresults in performance benefits. Additional exploration of employing Knowledge\nBase priors into the VAE reveals that the sentence space can be shifted towards\nthat of the Knowledge Base, offering interpretability and further improving\nresults.\n