Paper Link: https://openreview.net/forum?id=6_HFFqdYjSO
Paper Type: Long paper (up to eight pages of content + unlimited references and appendices)
Abstract: Extracting relational triples from unstructured text is crucial for information extraction. Recent methods extract relational triple from a stereoscopic perspective which can better capture the interaction between entity and relation. However, the stereoscopic models introduce redundant triples, which makes it difficult to identify triples accurately. Since the relation is one of the elements of triples to be extracted, the introduction of its semantic information can make the triple information more complete, which is helpful to relational triple extraction. In this work, we propose a Relation Semantic Information Attentive Stereoscopic framework (RSIA) which can fully represent and use the semantic information of relations. Specifically, a fusion encoder from transformers on top of relation encoder and sentence encoder is designed to enrich the semantic information of relation. Then, the semantic representation of the relation is integrated into the stereoscopic 3D space as its relation dimension. Our model achieves state-of-the-art performance with F1 score up to 93.5\% and 94.3\% on two public datasets and delivers consistent performance gain on complex scenarios of overlapping triples.