RECON: Relation Extraction using Knowledge Graph Context in a Graph Neural Network

In this paper, we present a novel method named RECON, that automatically\nidentifies relations in a sentence (sentential relation extraction) and aligns\nto a knowledge graph (KG). RECON uses a graph neural network to learn\nrepresentations of both the sentence as well as facts stored in a KG, improving\nthe overall extraction quality. These facts, including entity attributes\n(label, alias, description, instance-of) and factual triples, have not been\ncollectively used in the state of the art methods. We evaluate the effect of\nvarious forms of representing the KG context on the performance of RECON. The\nempirical evaluation on two standard relation extraction datasets shows that\nRECON significantly outperforms all state of the art methods on NYT Freebase\nand Wikidata datasets. RECON reports 87.23 F1 score (Vs 82.29 baseline) on\nWikidata dataset whereas on NYT Freebase, reported values are 87.5(P@10) and\n74.1(P@30) compared to the previous baseline scores of 81.3(P@10) and\n63.1(P@30).\n