Long distance entity relation extraction with article structure embedding and applied to mining medical knowledge

Yucong Lin, Cheng Ma, Daiqi Gao, Zihao Fan, Zijie Cheng, Zheyu Wang, Sheng Yu

Abstract: As a central work in medical knowledge graph construction, relation extraction has gained extensive attention in the fields of natural language processing and artificial intelligence. Conventional works on relation extraction share a common assumption: a sentence can express a relation of an entity pair only if both entities appear in this sentence. Under this assumption, plenty of informative sentences are precluded. In this paper, we break the assumption and propose a new relation extraction model that incorporates article structure information, which not only provide additional information, but also allows extracting long distance relations. We apply the model to online medical relation extraction and demonstrate its advantage over conventional models.