Chapter-level entity relationship extraction method based on joint learning
Jianqiong Xiao, Zhiyong Zhou · 2020
In this paper, we explored the relationship extraction on the coarse granularity of chapter-level, and proposed an entity relationship extraction method (CH-MEL) for chapter-level joint learning. First, the self-attention mechanism is used to learn the chapter representation based on the chapter structure information. Secondly, an improved joint annotation strategy is adopted to realize the annotation of entity relationship. Then, the joint decoding of entity and relationship is realized by introducing the minimize expected loss(MEL) training function. The experimental results in ACE05 and Semeval2010 Task 8 data sets show that the proposed joint relationship extraction method (Ch-MEL) is superior to the existing joint learning model.