KGGCN: Knowledge-Guided Graph Convolutional Networks for Distantly Supervised Relation Extraction

Distantly supervised relation extraction is the most popular technique for identifying semantic between two entities. Most prior models only focus on supervision information present in training sentences. In addition to sentences, external lexical resource and knowledge graphs often contain other relevant knowledge. However, usually ignore such readily available information. Moreover, previous works utilize a selective attention mechanism over sentences alleviate impact of noise, they lack consideration implicit interaction with facts. this paper, (1) knowledge-guided graph convolutional network proposed based word-level encode It can capture key words cue phrases generate expressive sentence-level features by attending indicators obtained from resource. (2) A sentence selector proposed, which explores structural triples as distinguish importance each individual sentence. Experimental results widely used datasets, NYT-FB GDS, show that our approach able efficiently use enhance performance distantly extraction.