Extracting Entities and Relations by an End-to-End Neural Model with Multi-Layer Encoder

Abstract

Extracting entities and relations in unstructured texts is an important task in the field of information extraction. However, the relations in sentences are often very complicated. Most existing methods mainly focus on the extraction of a single relation, and cannot accurately extract overlapping relations. In this paper, we propose an end-to-end neural model with multi-layer encoders to extract jointly overlapping relations and entities in sentences. In our model, we use character sequence of each word as supplementary information for word embedding, and adopt two bidirectional LSTMs to encode words and their character sequences to extract more of the semantic information in the text. Furthermore, we adopt multi-label classification to extract multiple relations. The difference from the major existing methods is that our model extracts the relations in sentences and combine each relation with encoding information to further extract entities. Finally, we conducted several experiments in the public dataset produced by distant supervision method to illustrate the performance of our method. The experimental results show that our model is superior to the previous methods, and our method has higher precision and F1 score.