A Moderately Deep Convolutional Neural Network for Relation Extraction

Abstract

Relation extraction in text data is considered as an important task in the field of natural language processing. So far, distant supervision is widely adopted in relation extraction to get labeled data. However, such a method is often lack of semantic information, and thus may bring wrong labelling problem. In this paper, a moderately deep convolutional neural network (CNN) is proposed to tackle the difficulty in relation extraction. The proposed CNN integrates low-level features of text sentences with high-level ones. The proposed CNN-based model has been evaluated on the NYT freebase larger dataset and the results show that our model is superior to the popular models such as CNN+ATT, PCNN+ATT, and ResCNN-9.