Joint Extraction of Entities and Overlapping Relations Using Position-Attentive Sequence Labeling

• DOI: 10.1609/aaai.v33i01.33016300
• Publication Date: 2019-08-27T07:40:07Z
• Authors: Dai Dai; Xinyan Xiao; Yajuan Lyu; Shan Dou; Qiaoqiao She; Haifeng Wang
• Keywords: Sequence labeling; Relationship extraction; Position (finance); Sequence (biology)

Joint entity and relation extraction is to detect using a single model. In this paper, we present novel unified joint model which directly tags labels according query word position p, i.e., detecting at identifying entities other positions that have relationship with the former. To end, first design tagging scheme generate n tag sequences for an n-word sentence. Then position-attention mechanism introduced produce different sentence representations every these sequences. way, our method can simultaneously extract all their type, as well overlapping relations. Experiment results show framework performances significantly better on extracting relations long-range relation, thus achieve state-of-the-art performance two public datasets.