LAUSR

Joint extraction of entities and relations is to recognize entities and semantic relations simultaneously, which is significant for knowledge graph construction. Recently, many effective joint models use dependency trees to capture the structural information of sentences. However, most dependency-based methods cannot make full use of the dependency information. This is because these methods just consider the connection information of dependency trees and ignore the influence of different nodes on the connected edges. In this paper, we establish a novel model to extract entities and relations simultaneously by using improved Multi-task Graph Convolutional networks, called MGCN. Specifically, considering the importance of node information, we merge both node and edge information into Graph Convolutional networks (GCN). In addition, in order to recognize the overlapping relations, we propose an efficient strategy to map multiple relational labels of a sentence into a unique code. Finally, we evaluate our joint model on two public datasets, and the experimental results show that our model outperforms the state-of-art models.