Joint Extraction of Entities and Relations via Entity and Relation Heterogeneous Graph Attention Networks

Entity and relation extraction (ERE) is a core task in information extraction. This has always faced the overlap problem. It was found that heterogeneous graph attention networks could enhance semantic analysis fusion between entities relations to improve ERE performance our previous work. In this paper, an entity network (ERHGA) proposed for joint ERE. A with gate mechanism constructed containing word nodes, subject nodes learn embedding of parts relational triple The ERHGA evaluated on public dataset named WebNLG. experimental results demonstrate ERHGA, by taking subjects as priori information, can effectively handle problem outperform all baselines 93.3%, especially overlapping triples.