Extraction Method with Word Distribution Enriched Deep Residual Network

Abstract

As a core task and important part of information extraction, relation extraction identifies the semantic relation between entity pairs. It plays an important role in semantic understanding of sentences and the construction of knowledge graphs. Most of the existing methods for relation extraction rely on semantic information. Furthermore, many word embedding models do not take position information into considerations. In this paper, combining with word vector representation of word embedding and words' positions, a word distribution model is proposed. It is used as the input of Residual Neural Network to train the classifier for relation extraction and Adversarial Training method is employed to reduce the impact of noise labels in training phase. The experimental results demonstrate the effectiveness of the proposed model on several datasets.