A Research on Overlapping Relationship Extraction Based on Multi-objective Dependency

Abstract

The joint extraction of entity and relation is an important task in information extraction. Previously, most models in entity relationship extraction assumed that the relationship was discrete. Unfortunately, this assumption is often violated. In order to solve the problem of overlapping in the entity relationship, considering the relationship between extraction under the premise of have the features of multiple targets, this paper puts forward a multi-objective depend on the relationship between extraction model, which transforms the relationship extraction task into a sequence-tagged task. The model uses Iterated Dilated Convolutional Neural Network (IDCNN) and BiLSTM to encode the words in order to more fully extract the semantics in the text. First, determine the target entity subject (s), and then predict all corresponding object (o) and relationship (r) according to s. Experiments show that our model is significantly better than the baseline methods.