Joint Extraction of Entities and Relations Using Reinforcement Learning and Deep Learning

Received Jan 23, 2017; Revised Apr 10, 2017; Accepted May 21, 2017

This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction

Information extraction [1] is the task of automatically extracting entities, relations, and events from unstructured texts. Researchers usually do research on entity extraction, relation extraction, and event extraction as separated tasks, but in fact there are important dependencies among tasks. For instance, entity information can further help relation extraction, so relation extraction takes the results of entity extraction as input. If just using a pipelined approach to tackle the above problem, information from each task cannot interact and get any feedback. Therefore, we make a detailed study of joint extraction of entities and relations from unstructured texts, which can pass the information of entity extraction to relation extraction and obtain feedback in order to improve the performance of entity extraction and relation extraction simultaneously.

In recent years, more and more researchers have applied deep learning to entity extraction and relation extraction. Huang et al. [2] proposed a bidirectional LSTM with a CRF layer (BILSTM-CRF) for sequence tagging, which included part-of-speech tagging (POS), chunking, and named entity recognition (NER). Nguyen and Grishman [3] proposed to combine the traditional feature-based method and the convolutional and recurrent neural networks for relation extraction. Deep learning can automatically extract features of entities and relations between entities to replace the method of designing features manually. It reduces the dependence of external resources and achieves good performance.

But how to pass entity information to relation extraction and obtain feedback is the research focus to the task of joint extraction of entities and relations, which means that we need an effective combination of different deep learning methods. To tackle the problem, we use reinforcement learning to model the task as a two-step decision process. Because it is difficult to find some measures to directly represent the state from unstructured texts, we use some deep...