Distantly Supervision for Relation Extraction via LayerNorm Gated Recurrent Neural Networks

Abstract

Relation extraction is a classic task in the NLP field which aims to predict the relation between two entities in given sentences. Convolutional neural network (CNN) is one of the typical neural network structures applied to this task. However, the existing CNN model used for extraction is not able to capture the time information in sentences which leads a great contribution to predict the right directionality between the two entities. Therefore, I propose a new gated recurrent neural networks with layer normalization (LNGRU) to obtain the background information of the future and the past in sentences. Experiments demonstrate that my model is effective and superior to several comparable baseline models.